demo

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+checkpoints/model.pth filter=lfs diff=lfs merge=lfs -text
+checkpoints/efficientnet.onnx filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+checkpoints/RawNet2.pth

Multimodal_deepfake_training_notebook.ipynb

@@ -0,0 +1,248 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "colab_type": "text",
+        "id": "view-in-github"
+      },
+      "source": [
+        "<a href=\"https://colab.research.google.com/github/AlvinKimata/ml-projects/blob/main/DFDT%20TMC/Multimodal_deepfake_training_notebook.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 1,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "FK1MZWm7oFa6",
+        "outputId": "ec19e080-086b-4cd6-997f-14dce5c61540"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Cloning into 'ml-projects'...\n",
+            "remote: Enumerating objects: 3730, done.\u001b[K\n",
+            "remote: Counting objects: 100% (719/719), done.\u001b[K\n",
+            "remote: Compressing objects: 100% (392/392), done.\u001b[K\n",
+            "remote: Total 3730 (delta 305), reused 710 (delta 298), pack-reused 3011\u001b[K\n",
+            "Receiving objects: 100% (3730/3730), 218.98 MiB | 9.61 MiB/s, done.\n",
+            "Resolving deltas: 100% (307/307), done.\n"
+          ]
+        }
+      ],
+      "source": [
+        "!git clone 'https://github.com/AlvinKimata/ml-projects.git'"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 2,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "IUb5rFqssg2j",
+        "outputId": "665d0e33-6d70-4873-d8ad-614dffdcf843"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "{\"username\":\"kaggle_username\",\"key\":\"kaggle_api_key\"}\n"
+          ]
+        }
+      ],
+      "source": [
+        "!mkdir ../root/.kaggle/\n",
+        "!echo '{\"username\":\"kaggle_username\",\"key\":\"kaggle_api_key\"}' >> /root/.kaggle/kaggle.json\n",
+        "!chmod 400 ../root/.kaggle/kaggle.json  #Read-only\n",
+        "!cat ../root/.kaggle/kaggle.json"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 4,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "owPZaNL8qAW8",
+        "outputId": "60e95755-df58-4906-e7ca-c9bb950c95cb"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Downloading fakeavceleb-tfrecord.zip to /content\n",
+            " 98% 1.52G/1.55G [00:20<00:00, 116MB/s]\n",
+            "100% 1.55G/1.55G [00:21<00:00, 79.2MB/s]\n"
+          ]
+        }
+      ],
+      "source": [
+        "!kaggle datasets download -d kimatadebonair/fakeavceleb-tfrecord"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 5,
+      "metadata": {
+        "id": "SG3kuPIJstaN"
+      },
+      "outputs": [],
+      "source": [
+        "!unzip -q '/content/fakeavceleb-tfrecord.zip' -d inputs/"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "CyAvPAhKgi9K"
+      },
+      "outputs": [],
+      "source": [
+        "!pip install -r 'DFDT TMC/requirements.txt'"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 6,
+      "metadata": {
+        "id": "sbBCy3Nps3V-"
+      },
+      "outputs": [],
+      "source": [
+        "!cp -r '/content/ml-projects/DFDT TMC' ./"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 8,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "LYmBafKPuGOM",
+        "outputId": "cebc40c2-40c2-4425-f4e8-55e7656df4d3"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "cp: cannot stat '/content/inputs/fakeavceleb_1k-000010-of-00015': No such file or directory\n",
+            "cp: cannot stat '/content/inputs/fakeavceleb_1k-000011-of-00015': No such file or directory\n",
+            "cp: cannot stat '/content/inputs/fakeavceleb_1k-000012-of-00015': No such file or directory\n",
+            "cp: cannot stat '/content/inputs/fakeavceleb_1k-000013-of-00015': No such file or directory\n"
+          ]
+        }
+      ],
+      "source": [
+        "for i in range(14):\n",
+        "  !cp '/content/inputs/fakeavceleb_1k-0000{i}-of-00015' '/content/DFDT TMC/datasets/train'"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 11,
+      "metadata": {
+        "id": "O1mT677Uc0qu"
+      },
+      "outputs": [],
+      "source": [
+        "for i in range(10, 15):\n",
+        "  !cp '/content/inputs/fakeavceleb_1k-000{i}-of-00015' '/content/DFDT TMC/datasets/train'"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 32,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "_-TCpjHVqT36",
+        "outputId": "88873108-392c-4830-f4e4-76b3a2cc8b3c"
+      },
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "--2023-07-14 09:10:01--  https://github.com/selimsef/dfdc_deepfake_challenge/releases/download/0.0.1/final_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23\n",
+            "Resolving github.com (github.com)... 192.30.255.112\n",
+            "Connecting to github.com (github.com)|192.30.255.112|:443... connected.\n",
+            "HTTP request sent, awaiting response... 302 Found\n",
+            "Location: https://objects.githubusercontent.com/github-production-release-asset-2e65be/270020698/6e91bf80-a835-11ea-8950-51c980e899ce?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAIWNJYAX4CSVEH53A%2F20230714%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20230714T091002Z&X-Amz-Expires=300&X-Amz-Signature=8623af355287f61ac5b0e7857ae8c21efdbeb265ccc3662b57cee5f04f31f572&X-Amz-SignedHeaders=host&actor_id=0&key_id=0&repo_id=270020698&response-content-disposition=attachment%3B%20filename%3Dfinal_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23&response-content-type=application%2Foctet-stream [following]\n",
+            "--2023-07-14 09:10:02--  https://objects.githubusercontent.com/github-production-release-asset-2e65be/270020698/6e91bf80-a835-11ea-8950-51c980e899ce?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAIWNJYAX4CSVEH53A%2F20230714%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20230714T091002Z&X-Amz-Expires=300&X-Amz-Signature=8623af355287f61ac5b0e7857ae8c21efdbeb265ccc3662b57cee5f04f31f572&X-Amz-SignedHeaders=host&actor_id=0&key_id=0&repo_id=270020698&response-content-disposition=attachment%3B%20filename%3Dfinal_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23&response-content-type=application%2Foctet-stream\n",
+            "Resolving objects.githubusercontent.com (objects.githubusercontent.com)... 185.199.111.133, 185.199.109.133, 185.199.110.133, ...\n",
+            "Connecting to objects.githubusercontent.com (objects.githubusercontent.com)|185.199.111.133|:443... connected.\n",
+            "HTTP request sent, awaiting response... 200 OK\n",
+            "Length: 266910615 (255M) [application/octet-stream]\n",
+            "Saving to: ‘final_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23’\n",
+            "\n",
+            "final_999_DeepFakeC 100%[===================>] 254.54M  66.8MB/s    in 3.8s    \n",
+            "\n",
+            "2023-07-14 09:10:06 (66.4 MB/s) - ‘final_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23’ saved [266910615/266910615]\n",
+            "\n"
+          ]
+        }
+      ],
+      "source": [
+        "!cd '/content/DFDT TMC/pretrained' && wget 'https://github.com/selimsef/dfdc_deepfake_challenge/releases/download/0.0.1/final_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23'''"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "DvA-myf8s9-9",
+        "outputId": "477f4488-e1fb-44bb-b867-71c325c85dcb"
+      },
+      "outputs": [],
+      "source": [
+        "!python '/content/DFDT TMC/train_dfdc_tf.py' --device='cuda' \\\n",
+        "  --data_dir=\"/content/DFDT TMC/datasets/train/fakeavceleb_1k*\" \\\n",
+        "  --pretrained_image_encoder=True --pretrained_audio_encoder=True"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "kGfym7pEn4aP"
+      },
+      "outputs": [],
+      "source": []
+    }
+  ],
+  "metadata": {
+    "accelerator": "GPU",
+    "colab": {
+      "authorship_tag": "ABX9TyNzEVTklkrYn6Mgz+yxoZaI",
+      "gpuType": "T4",
+      "include_colab_link": true,
+      "provenance": []
+    },
+    "kernelspec": {
+      "display_name": "Python 3",
+      "name": "python3"
+    },
+    "language_info": {
+      "name": "python"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

app.py

@@ -0,0 +1,35 @@
+import gradio as gr
+import inference_2 as inference
+
+
+title="Multimodal deepfake detector"
+description="Deepfake detection for videos, images and audio modalities."
+            
+           
+video_interface = gr.Interface(inference.deepfakes_video_predict,
+                    gr.Video(),
+                    "text",
+                    examples = ["videos/celeb_synthesis.mp4", "videos/real-1.mp4"],
+                    cache_examples = False
+                    )
+
+
+image_interface = gr.Interface(inference.deepfakes_image_predict,
+                    gr.Image(),
+                    "text",
+                    examples = ["images/lady.jpg", "images/fake_image.jpg"],
+                    cache_examples=False
+                    )
+
+audio_interface = gr.Interface(inference.deepfakes_spec_predict,
+                               gr.Audio(),
+                               "text",
+                               examples = ["audios/DF_E_2000027.flac", "audios/DF_E_2000031.flac"],
+                               cache_examples = False)
+
+
+app = gr.TabbedInterface(interface_list= [image_interface, video_interface, audio_interface], 
+                         tab_names = ['Image inference', 'Video inference', 'Audio inference'])
+
+if __name__ == '__main__':
+    app.launch(share = False)

checkpoints/efficientnet.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:206f99f4c4efe6d088ba6e53bfcdec76ffa796a345d50770c037005e3cd11639
+size 23510323

checkpoints/model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3de812710093068acee6200b8d162aab074975edffa3edf2ccbe562868e4adf6
+size 117418889

data/__init__.py

@@ -0,0 +1,22 @@
+import torch.utils.data
+
+class DataProvider():
+
+    def __init__(self, cfg, dataset, batch_size=None, shuffle=True):
+        super().__init__()
+        self.dataset = dataset
+        if batch_size is None:
+            batch_size = cfg.BATCH_SIZE
+        self.dataloader = torch.utils.data.DataLoader(
+            self.dataset,
+            batch_size=batch_size,
+            shuffle=shuffle,
+            num_workers=int(cfg.WORKERS),
+            drop_last=False)
+
+    def __len__(self):
+        return len(self.dataset)
+
+    def __iter__(self):
+        for i, data in enumerate(self.dataloader):
+            yield data

data/augmentation_utils.py

@@ -0,0 +1,88 @@
+import cv2
+import librosa
+import numpy as np
+import albumentations
+from albumentations import (Compose, ImageCompression, GaussNoise, HorizontalFlip, 
+                            PadIfNeeded, OneOf,ToGray, ShiftScaleRotate, GaussianBlur,
+                            RandomBrightnessContrast, FancyPCA, HueSaturationValue, BasicTransform)
+
+
+class AudioTransform(BasicTransform):
+    """ Transform for audio task. This is the main class where we override the targets and update params function for our need"""
+    @property
+    def targets(self):
+        return {"data": self.apply}
+    
+    def update_params(self, params, **kwargs):
+        if hasattr(self, "interpolation"):
+            params["interpolation"] = self.interpolation
+        if hasattr(self, "fill_value"):
+            params["fill_value"] = self.fill_value
+        return params
+
+class TimeShifting(AudioTransform):
+    """ Do time shifting of audio """
+    def __init__(self, always_apply=False, p=0.5):
+        super(TimeShifting, self).__init__(always_apply, p)
+        
+    def apply(self,data,**params):
+        '''
+        data : ndarray of audio timeseries
+        '''        
+        start_ = int(np.random.uniform(-80000,80000))
+        if start_ >= 0:
+            audio_time_shift = np.r_[data[start_:], np.random.uniform(-0.001,0.001, start_)]
+        else:
+            audio_time_shift = np.r_[np.random.uniform(-0.001,0.001, -start_), data[:start_]]
+        
+        return audio_time_shift
+
+class PitchShift(AudioTransform):
+    """ Do time shifting of audio """
+    def __init__(self, always_apply=False, p=0.5 , n_steps=None):
+        super(PitchShift, self).__init__(always_apply, p)
+        '''
+        nsteps here is equal to number of semitones
+        '''
+        
+        self.n_steps = n_steps
+        
+    def apply(self,data,**params):
+        '''
+        data : ndarray of audio timeseries
+        '''        
+        return librosa.effects.pitch_shift(data,sr=16000,n_steps=self.n_steps)
+
+
+class AddGaussianNoise(AudioTransform):
+    """ Do time shifting of audio """
+    def __init__(self, always_apply=False, p=0.5):
+        super(AddGaussianNoise, self).__init__(always_apply, p)
+        
+        
+    def apply(self,data,**params):
+        '''
+        data : ndarray of audio timeseries
+        ''' 
+        noise = np.random.randn(len(data))
+        data_wn = data + 0.005*noise
+        return data_wn
+
+
+create_frame_transforms = Compose([
+        ImageCompression(quality_lower=60, quality_upper=100, p=0.5),
+        GaussNoise(p=0.1),
+        GaussianBlur(blur_limit=3, p=0.05),
+        HorizontalFlip(),
+        PadIfNeeded(min_height=256, min_width=256, border_mode=cv2.BORDER_CONSTANT),
+        OneOf([RandomBrightnessContrast(), FancyPCA(), HueSaturationValue()], p=0.7),
+        ToGray(p=0.2),
+        ShiftScaleRotate(shift_limit=0.1, scale_limit=0.2, rotate_limit=10, border_mode=cv2.BORDER_CONSTANT, p=0.5),])
+
+
+
+create_spec_transforms = albumentations.Compose([
+        TimeShifting(p=0.9),  # here not p=1.0 because your nets should get some difficulties
+        AddGaussianNoise(p=0.8),
+        PitchShift(p=0.5,n_steps=4)
+    ])

data/dfdt_dataset.py

@@ -0,0 +1,130 @@
+'''Module for loading the fakeavceleb dataset from tfrecord format'''
+import numpy as np
+import tensorflow as tf
+from data.augmentation_utils import create_frame_transforms, create_spec_transforms
+
+FEATURE_DESCRIPTION = {
+    'video_path': tf.io.FixedLenFeature([], tf.string), 
+    'image/encoded': tf.io.FixedLenFeature([], tf.string),
+    'clip/label/index': tf.io.FixedLenFeature([], tf.int64),
+    'clip/label/text': tf.io.FixedLenFeature([], tf.string), 
+    'WAVEFORM/feature/floats': tf.io.FixedLenFeature([], tf.string)
+}
+
+@tf.function
+def _parse_function(example_proto):
+
+    #Parse the input `tf.train.Example` proto using the dictionary above.
+    example = tf.io.parse_single_example(example_proto, FEATURE_DESCRIPTION)
+    
+    video_path = example['video_path']
+    video = tf.io.decode_raw(example['image/encoded'], tf.int8)    
+    spectrogram = tf.io.decode_raw(example['WAVEFORM/feature/floats'], tf.float32)
+    
+    label = example["clip/label/text"]
+    label_map = example["clip/label/index"]
+    
+    return video, spectrogram, label_map
+
+@tf.function
+def decode_inputs(video, spectrogram, label_map):
+    '''Decode tensors to arrays with desired shape'''
+    frame = tf.reshape(video, [10, 3, 256, 256])
+    frame = frame[0] / 255 #Pick the first frame and normalize it.
+    # frame = tf.cast(frame, tf.float32)
+
+    label_map = tf.expand_dims(label_map, axis = 0)
+    
+    sample = {'video_reshaped': frame, 'spectrogram': spectrogram, 'label_map': label_map}
+    return sample
+
+
+def decode_train_inputs(video, spectrogram, label_map):
+    #Data augmentation for spectograms
+    spectrogram_shape = spectrogram.shape
+    spec_augmented = tf.py_function(aug_spec_fn, [spectrogram], tf.float32)
+    spec_augmented.set_shape(spectrogram_shape)
+
+    frame = tf.reshape(video, [10, 256, 256, 3])
+    frame = frame[0] #Pick the first frame.
+    frame = frame / 255 #Normalize tensor.
+
+    frame_augmented = tf.py_function(aug_img_fn, [frame], tf.uint8)
+    # frame_augmented.set_shape(frame_shape)
+
+    frame_augmented.set_shape([3, 256, 256])
+    label_map = tf.expand_dims(label_map, axis = 0)
+
+    augmented_sample = {'video_reshaped': frame_augmented, 'spectrogram': spec_augmented, 'label_map': label_map}
+    return augmented_sample
+
+
+def aug_img_fn(frame):
+  frame = frame.numpy().astype(np.uint8)
+  frame_data = {'image': frame}
+  aug_frame_data = create_frame_transforms(**frame_data)
+  aug_img = aug_frame_data['image']
+  aug_img = aug_img.transpose(2, 0, 1)
+  return aug_img
+
+def aug_spec_fn(spec):
+  spec = spec.numpy()
+  spec_data = {'spec': spec}
+  aug_spec_data = create_spec_transforms(**spec_data)
+  aug_spec = aug_spec_data['spec']
+  return aug_spec
+
+
+class FakeAVCelebDatasetTrain:
+
+    def __init__(self, args):
+        self.args = args
+        self.samples = self.load_features_from_tfrec()
+
+    def load_features_from_tfrec(self):
+        '''Loads raw features from a tfrecord file and returns them as raw inputs'''
+        ds = tf.io.matching_files(self.args.data_dir)
+        files = tf.random.shuffle(ds)
+
+        shards = tf.data.Dataset.from_tensor_slices(files)
+        dataset = shards.interleave(tf.data.TFRecordDataset)
+        dataset = dataset.shuffle(buffer_size=100)
+
+        dataset = dataset.map(_parse_function, num_parallel_calls = tf.data.AUTOTUNE)
+        dataset = dataset.map(decode_train_inputs, num_parallel_calls = tf.data.AUTOTUNE)
+        dataset = dataset.padded_batch(batch_size = self.args.batch_size)
+        return dataset
+    
+    
+    def __len__(self):
+        self.samples = self.load_features_from_tfrec(self.args.data_dir)
+        cnt = self.samples.reduce(np.int64(0), lambda x, _: x + 1)
+        cnt = cnt.numpy()
+        return cnt
+
+class FakeAVCelebDatasetVal:
+
+    def __init__(self, args):
+        self.args = args
+        self.samples = self.load_features_from_tfrec()
+
+    def load_features_from_tfrec(self):
+        '''Loads raw features from a tfrecord file and returns them as raw inputs'''
+        ds = tf.io.matching_files(self.args.data_dir)
+        files = tf.random.shuffle(ds)
+
+        shards = tf.data.Dataset.from_tensor_slices(files)
+        dataset = shards.interleave(tf.data.TFRecordDataset)
+        dataset = dataset.shuffle(buffer_size=100)
+
+        dataset = dataset.map(_parse_function, num_parallel_calls = tf.data.AUTOTUNE)
+        dataset = dataset.map(decode_inputs, num_parallel_calls = tf.data.AUTOTUNE)
+        dataset = dataset.padded_batch(batch_size = self.args.batch_size)
+        return dataset
+    
+    
+    def __len__(self):
+        self.samples = self.load_features_from_tfrec(self.args.data_dir)
+        cnt = self.samples.reduce(np.int64(0), lambda x, _: x + 1)
+        cnt = cnt.numpy()
+        return cnt

data/generate_dataset_to_tfrecord.py

@@ -0,0 +1,178 @@
+#Code outsourced from https://github.com/deepmind/dmvr/tree/master and later modified.
+
+"""Python script to generate TFRecords of SequenceExample from raw videos."""
+
+import contextlib
+import math
+import os
+import cv2
+from typing import Dict, Optional, Sequence
+import moviepy.editor
+from absl import app
+from absl import flags
+import ffmpeg
+import numpy as np
+import pandas as pd
+import tensorflow as tf
+
+import warnings
+warnings.filterwarnings('ignore')
+
+flags.DEFINE_string("csv_path", "fakeavceleb_1k.csv", "Input csv")
+flags.DEFINE_string("output_path", "fakeavceleb_tfrec", "Tfrecords output path.")
+flags.DEFINE_string("video_root_path", "./",
+                    "Root directory containing the raw videos.")
+flags.DEFINE_integer(
+    "num_shards", 4, "Number of shards to output, -1 means"
+    "it will automatically adapt to the sqrt(num_examples).")
+flags.DEFINE_bool("decode_audio", False, "Whether or not to decode the audio")
+flags.DEFINE_bool("shuffle_csv", False, "Whether or not to shuffle the csv.")
+FLAGS = flags.FLAGS
+
+
+_JPEG_HEADER = b"\xff\xd8"
+
+
+@contextlib.contextmanager
+def _close_on_exit(writers):
+  """Call close on all writers on exit."""
+  try:
+    yield writers
+  finally:
+    for writer in writers:
+      writer.close()
+
+
+def add_float_list(key: str, values: Sequence[float],
+                   sequence: tf.train.SequenceExample):
+  sequence.feature_lists.feature_list[key].feature.add(
+  ).float_list.value[:] = values
+
+
+def add_bytes_list(key: str, values: Sequence[bytes],
+                   sequence: tf.train.SequenceExample):
+  sequence.feature_lists.feature_list[key].feature.add().bytes_list.value[:] = values
+
+
+def add_int_list(key: str, values: Sequence[int],
+                 sequence: tf.train.SequenceExample):
+  sequence.feature_lists.feature_list[key].feature.add().int64_list.value[:] = values
+
+
+def set_context_int_list(key: str, value: Sequence[int],
+                         sequence: tf.train.SequenceExample):
+  sequence.context.feature[key].int64_list.value[:] = value
+
+
+def set_context_bytes(key: str, value: bytes,
+                      sequence: tf.train.SequenceExample):
+  sequence.context.feature[key].bytes_list.value[:] = (value,)
+
+def set_context_bytes_list(key: str, value: Sequence[bytes],
+                           sequence: tf.train.SequenceExample):
+   sequence.context.feature[key].bytes_list.value[:] = value
+
+
+def set_context_float(key: str, value: float,
+                      sequence: tf.train.SequenceExample):
+  sequence.context.feature[key].float_list.value[:] = (value,)
+
+
+def set_context_int(key: str, value: int, sequence: tf.train.SequenceExample):
+  sequence.context.feature[key].int64_list.value[:] = (value,)
+
+
+def extract_frames(video_path, fps = 10, min_resize = 256):
+    '''Load n number of frames from a video'''
+    v_cap = cv2.VideoCapture(video_path)
+    v_len = int(v_cap.get(cv2.CAP_PROP_FRAME_COUNT))
+    
+    if fps is None:
+        sample = np.arange(0, v_len)
+    else:
+        sample = np.linspace(0, v_len - 1, fps).astype(int)
+
+    frames = []
+    for j in range(v_len):
+        success = v_cap.grab()
+        if j in sample:
+            success, frame = v_cap.retrieve()
+            if not success:
+                continue
+                
+            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+            frame = cv2.resize(frame, (min_resize, min_resize))
+            frames.append(frame)
+    
+    v_cap.release()
+    frame_np = np.stack(frames)
+    return frame_np.tobytes()
+
+def extract_audio(video_path: str,
+                  sampling_rate: int = 16_000):
+  """Extract raw mono audio float list from video_path with ffmpeg."""
+  video = moviepy.editor.VideoFileClip(video_path)
+  audio = video.audio.to_soundarray()
+  #Load first channel.
+  audio = audio[:, 0]
+  
+  return np.array(audio)
+
+#Each of the features can be coerced into a tf.train.Example-compatible type using one of the _bytes_feature, _float_feature and the _int64_feature.
+#You can then create a tf.train.Example message from these encoded features.
+
+def serialize_example(video_path: str, label_name: str, label_map: Optional[Dict[str, int]] = None):
+    # Initiate the sequence example.
+    seq_example = tf.train.SequenceExample()
+
+    imgs_encoded = extract_frames(video_path, fps = 10)
+
+    audio = extract_audio(video_path)
+
+    set_context_bytes(f'image/encoded', imgs_encoded, seq_example)
+    set_context_bytes("video_path", video_path.encode(), seq_example)
+    set_context_bytes("WAVEFORM/feature/floats", audio.tobytes(), seq_example)
+    set_context_int("clip/label/index", label_map[label_name], seq_example)
+    set_context_bytes("clip/label/text", label_name.encode(), seq_example)
+    return seq_example
+
+
+def main(argv):
+    del argv
+    # reads the input csv.
+    input_csv = pd.read_csv(FLAGS.csv_path)
+    if FLAGS.num_shards == -1:
+        num_shards = int(math.sqrt(len(input_csv)))
+    else:
+        num_shards = FLAGS.num_shards
+    # Set up the TFRecordWriters.
+    basename = os.path.splitext(os.path.basename(FLAGS.csv_path))[0]
+    shard_names = [
+      os.path.join(FLAGS.output_path, f"{basename}-{i:05d}-of-{num_shards:05d}")
+      for i in range(num_shards)
+    ]
+    writers = [tf.io.TFRecordWriter(shard_name) for shard_name in shard_names]
+
+    if "label" in input_csv:
+        unique_labels = list(set(input_csv["label"].values))
+        l_map = {unique_labels[i]: i for i in range(len(unique_labels))}
+    else:
+        l_map = None
+
+    if FLAGS.shuffle_csv:
+        input_csv = input_csv.sample(frac=1)
+    with _close_on_exit(writers) as writers:
+        row_count = 0
+        for row in input_csv.itertuples():
+           index = row[0]
+           v = row[1]
+           if os.name == 'posix':
+            v = v.str.replace('\\', '/')
+           l = row[2]
+           row_count += 1
+           print("Processing example %d of %d   (%d%%) \r" %(row_count, len(input_csv), row_count * 100 / len(input_csv)), end="")
+           seq_ex = serialize_example(video_path = v, label_name = l,label_map = l_map)
+           writers[index % len(writers)].write(seq_ex.SerializeToString())
+
+if __name__ == "__main__":
+  app.run(main)

datasets/fakeavceleb_100.csv

@@ -0,0 +1,101 @@
+video_path,label
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00076/00109.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00166/00010.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00173/00118.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00366/00118.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00391/00052.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00475/00099.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00476/00109.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00478/00206.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00518/00031.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00701/00092.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00761/00072.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00781/00092.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00830/00143.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00944/00135.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id00987/00160.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01036/00010.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01076/00005.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01170/00021.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01171/00053.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01179/00160.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01207/00320.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01236/00005.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01392/00167.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01452/00001.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01521/00109.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01528/00017.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01530/00002.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01544/00044.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01597/00005.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01598/00044.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01610/00090.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01637/00002.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01691/00045.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01717/00005.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01779/00010.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01835/00130.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01856/00006.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01920/00099.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01933/00028.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01972/00078.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id01995/00071.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02005/00052.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02040/00476.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02051/00015.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02268/00036.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02296/00019.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02316/00094.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02342/00191.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id02494/00050.mp4,real
+FakeAVCeleb/RealVideo-RealAudio/African/men/id04727/00007.mp4,real
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_10_id00476_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_10_id01076_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_10_id01179_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_10_id02005_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_10_id02342_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_12_id00518_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_12_id00761_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_12_id00987_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_12_id01856_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_12_id02296_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_2_id00166_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_2_id00701_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_2_id01236_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_2_id01521_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_2_id01598_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_4_id01392_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_4_id01528_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_4_id01691_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_4_id01995_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_4_id02296_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_7_id00166_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_7_id00478_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_7_id01452_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_7_id01717_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_7_id01995_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_8_id00166_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_8_id00701_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_8_id00761_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_8_id01170_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_8_id02005_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_9_id00076_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_9_id01036_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_9_id01452_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_9_id01528_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00076/00109_9_id02005_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00166/00010_id01637_5VjcPZm8knM_faceswap_id00166_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00166/00010_id01637_5VjcPZm8knM_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00166/00010_id01637_5VjcPZm8knM_faceswap_id01171_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00166/00010_id01637_5VjcPZm8knM_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00166/00010_id01637_5VjcPZm8knM_faceswap_id01598_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00173/00118_id00476_UgdYVJ6xPYg_faceswap_id00166_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00173/00118_id00476_UgdYVJ6xPYg_faceswap_id00173_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00173/00118_id00476_UgdYVJ6xPYg_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00173/00118_id00476_UgdYVJ6xPYg_faceswap_id01598_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00173/00118_id00476_UgdYVJ6xPYg_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00366/00118_id00076_Isiq7cA-DNE_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00366/00118_id00076_Isiq7cA-DNE_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00366/00118_id00076_Isiq7cA-DNE_faceswap_id02316_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00366/00118_id00076_Isiq7cA-DNE_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb/FakeVideo-FakeAudio/African/men/id00366/00118_id00076_Isiq7cA-DNE_faceswap_id02494_wavtolip.mp4,fake

datasets/fakeavceleb_1k.csv

@@ -0,0 +1,1001 @@
+video_path,label
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00076\00109.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00166\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00173\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00366\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00391\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00475\00099.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00476\00109.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00478\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00518\00031.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00701\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00761\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00781\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00830\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00944\00135.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id00987\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01036\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01076\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01170\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01171\00053.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01179\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01207\00320.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01236\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01392\00167.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01452\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01521\00109.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01528\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01530\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01544\00044.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01597\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01598\00044.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01610\00090.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01637\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01691\00045.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01717\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01779\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01835\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01856\00006.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01920\00099.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01933\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01972\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id01995\00071.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02005\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02040\00476.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02051\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02268\00036.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02296\00019.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02316\00094.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02342\00191.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id02494\00050.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\men\id04727\00007.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00220\00027.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00359\00053.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00371\00099.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00460\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00568\00384.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00577\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00592\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00707\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00829\00271.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id00832\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01178\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01532\00065.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01661\00059.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01783\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01838\00126.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id01907\00148.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02071\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02301\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02508\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02586\00042.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02617\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02721\00424.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02808\00056.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02824\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02838\00080.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id02948\00298.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03103\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03569\00065.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03656\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03658\00077.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03713\00249.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id03747\00273.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04055\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04245\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04374\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04376\00181.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04437\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04540\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04547\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04689\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04705\00408.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04736\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04820\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id04939\00174.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05106\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05231\00149.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05235\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05251\00033.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05252\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\African\women\id05980\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00018\00181.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00020\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00021\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00029\00288.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00049\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00052\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00060\00307.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00062\00278.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00087\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00088\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00169\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00179\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00184\00241.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00243\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00264\00257.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00345\00243.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00529\00409.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00696\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00708\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00775\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00777\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00943\00304.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00945\00107.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id00971\00253.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01035\00012.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01042\00154.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01044\00336.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01048\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01096\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01105\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01124\00063.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01163\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01168\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01172\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01175\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01182\00167.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01192\00217.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01201\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01210\00283.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01211\00023.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id01239\00280.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id03525\00048.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id03668\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id03678\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id03757\00149.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id04034\00009.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id04073\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id04216\00470.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id04219\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\men\id04221\00053.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00025\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00097\00162.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00098\00004.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00100\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00145\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00180\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00190\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00231\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00261\00048.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00272\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00291\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00381\00030.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00383\00171.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00385\00439.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00398\00016.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00418\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00428\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00431\00039.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00458\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00462\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00555\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00575\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00616\00305.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00618\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00680\00110.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00752\00340.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00835\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00841\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00842\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id00848\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01004\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01005\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01075\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01091\00236.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01216\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01217\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01223\00255.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01225\00300.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01227\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01231\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01238\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01245\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id01248\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id02464\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id02466\00136.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id03556\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id03605\00048.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id03696\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id03707\00055.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (American)\women\id03781\00113.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00056\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00126\00173.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00560\00041.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00597\00019.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00740\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id00863\00069.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id01204\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id01212\00183.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id01215\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id01589\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id01683\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id02332\00055.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id02365\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id02493\00073.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id02553\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id02561\02561.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id03028\00466.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id03168\03168.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id03889\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id03965\00051.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04111\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04222\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04687\00066.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04691\1.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04726\00245.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04774\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04789\002121.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id04884\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id05268\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id05332\00065.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id05383\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id05479\05479.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id05743\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06152\06152.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06269\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06467\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06470\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06535\00183.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06591\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06594\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06776\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06807\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id06878\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id07102\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id07338\00003.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id08299\00110.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id08613\00074.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id08652\00006.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id09053\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\men\id09143\00056.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00137\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00363\00014.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00430\00209.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00566\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00579\00030.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00582\00006.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00763\00074.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id00935\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id01281\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id01451\00099.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id02587\00020.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id02807\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id03211\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id03379\00032.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id03940\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id04057\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id04066\00013.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id04144\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id04414\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id04701\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id05576\00368.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id05620\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id05631\00073.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id05844\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06054\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06060\00219.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06061\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06065\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06066\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06158\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06225\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06388\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06427\00138.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06443\00232.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id06462\00014.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id07039\00105.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id07383\00011.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id07739\00019.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id07799\00063.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id07901\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id08139\00067.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id08397\00167.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id08402\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id08819\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09116\00026.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09125\00098.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09171\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09174\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09175\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (East)\women\id09181\00048.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00055\00120.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00063\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00171\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00183\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00185\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00186\00120.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00187\00360.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00192\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00225\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00241\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00253\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00265\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00266\00470.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00282\00268.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00292\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00305\00113.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00306\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00358\00217.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00368\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00415\00017.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00498\00014.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00519\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00520\00187.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00535\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00548\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00554\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00559\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00594\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00694\00340.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00709\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00909\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00919\00063.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00946\00126.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00963\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00981\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00990\00160.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id00999\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01051\00322.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01052\00076.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01058\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01098\00044.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01099\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01102\00197.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01123\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01126\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01154\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01156\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id01157\00048.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id02567\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\men\id03205\00150.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00042\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00068\00004.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00071\00014.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00188\00020.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00232\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00234\00063.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00262\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00270\00088.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00271\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00287\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00325\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00328\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00330\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00332\00293.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00365\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00373\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00374\00311.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00379\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00395\00420.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00403\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00434\00046.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00484\00202.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00491\00122.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00495\00027.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00496\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00569\00239.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00581\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00589\00130.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00591\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00633\00088.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00634\00078.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00735\00037.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00806\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00813\00169.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00823\00125.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00826\00065.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id00897\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id01001\00086.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id01002\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id01018\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03371\00430.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03589\00002.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03620\00081.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03649\00001.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03651\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03716\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03816\00093.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03844\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03858\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Caucasian (European)\women\id03941\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00032\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00033\00276.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00078\00114.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00082\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00103\00241.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00350\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00414\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00459\00382.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00685\00146.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00732\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00745\00165.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00769\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00773\00038.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00816\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00857\00347.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id00860\00154.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id03180\00039.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id03344\00114.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id03599\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id03945\00063.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04526\00317.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04537\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04554\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04560\00195.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04561\00248.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04562\00221.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04599\00111.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04601\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id04928\00027.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id06334\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id06354\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id06355\00347.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id06753\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07058\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07108\00412.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07161\00159.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07163\00141.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07165\00368.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07179\00206.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07182\00040.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07194\00014.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07195\00186.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07200\00045.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07210\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07233\00010.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07463\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id07768\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id08313\00202.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id08314\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\men\id08457\00417.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00043\00135.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00080\00281.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00149\00284.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00235\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00417\00069.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00461\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00488\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00739\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id00747\00053.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id01026\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id02089\00092.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id02310\00139.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id02619\00015.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id03559\00023.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id03815\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id03897\00021.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id03985\00005.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04070\00072.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04490\00054.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04529\00186.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04530\00231.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04564\00417.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04582\00180.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04583\00077.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id04927\00013.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05434\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05435\00107.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05478\00135.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05845\00027.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05920\00161.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id05931\00013.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06232\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06254\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06268\00159.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06343\00023.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06428\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06437\00028.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06438\00110.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06439\00118.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06445\00150.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id06752\00221.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07008\00175.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07049\00043.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07051\00083.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07078\00405.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07136\00052.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07236\00143.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07377\00025.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07686\00254.mp4,real
+FakeAVCeleb\RealVideo-RealAudio\Asian (South)\women\id07689\00028.mp4,real
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_10_id00476_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_10_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_10_id01179_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_10_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_10_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_12_id00518_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_12_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_12_id00987_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_12_id01856_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_12_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_2_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_2_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_2_id01236_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_2_id01521_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_2_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_4_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_4_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_4_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_4_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_4_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_7_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_7_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_7_id01452_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_7_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_7_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_8_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_8_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_8_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_8_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_8_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_9_id00076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_9_id01036_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_9_id01452_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_9_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00076\00109_9_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00166\00010_id01637_5VjcPZm8knM_faceswap_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00166\00010_id01637_5VjcPZm8knM_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00166\00010_id01637_5VjcPZm8knM_faceswap_id01171_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00166\00010_id01637_5VjcPZm8knM_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00166\00010_id01637_5VjcPZm8knM_faceswap_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00173\00118_id00476_UgdYVJ6xPYg_faceswap_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00173\00118_id00476_UgdYVJ6xPYg_faceswap_id00173_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00173\00118_id00476_UgdYVJ6xPYg_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00173\00118_id00476_UgdYVJ6xPYg_faceswap_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00173\00118_id00476_UgdYVJ6xPYg_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00366\00118_id00076_Isiq7cA-DNE_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00366\00118_id00076_Isiq7cA-DNE_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00366\00118_id00076_Isiq7cA-DNE_faceswap_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00366\00118_id00076_Isiq7cA-DNE_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00366\00118_id00076_Isiq7cA-DNE_faceswap_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00391\00052_id00476_UgdYVJ6xPYg_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00391\00052_id00476_UgdYVJ6xPYg_faceswap_id01179_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00391\00052_id00476_UgdYVJ6xPYg_faceswap_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00391\00052_id00476_UgdYVJ6xPYg_faceswap_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00391\00052_id00476_UgdYVJ6xPYg_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00475\00099_0_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00475\00099_0_id01920_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00475\00099_0_id01972_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00475\00099_0_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00475\00099_0_id04727_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00476_UgdYVJ6xPYg_faceswap_id00076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00476_UgdYVJ6xPYg_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00476_UgdYVJ6xPYg_faceswap_id00781_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00476_UgdYVJ6xPYg_faceswap_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00476_UgdYVJ6xPYg_faceswap_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00781_fvsSae9yc0A_faceswap_id00476_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00781_fvsSae9yc0A_faceswap_id00944_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00781_fvsSae9yc0A_faceswap_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00781_fvsSae9yc0A_faceswap_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00476\00109_id00781_fvsSae9yc0A_faceswap_id04727_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00478\00109_11_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00478\00109_11_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00478\00109_11_id01856_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00478\00109_11_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00478\00109_11_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_0_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_0_id00391_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_0_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_0_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_0_id02268_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_1_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_1_id00987_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_1_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_1_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_1_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_2_id01544_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_2_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_2_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_2_id01835_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_2_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_3_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_3_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_3_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_3_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00518\00031_3_id02268_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00701\00092_id01036_AohKaMtIHxA_faceswap_id00391_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00701\00092_id01036_AohKaMtIHxA_faceswap_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00701\00092_id01036_AohKaMtIHxA_faceswap_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00701\00092_id01036_AohKaMtIHxA_faceswap_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00701\00092_id01036_AohKaMtIHxA_faceswap_id01972_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00761\00072_id01835_UZbWA0QfXXA_faceswap_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00761\00072_id01835_UZbWA0QfXXA_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00761\00072_id01835_UZbWA0QfXXA_faceswap_id01036_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00761\00072_id01835_UZbWA0QfXXA_faceswap_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00761\00072_id01835_UZbWA0QfXXA_faceswap_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00781\00092_id00476_UgdYVJ6xPYg_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00781\00092_id00476_UgdYVJ6xPYg_faceswap_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00781\00092_id00476_UgdYVJ6xPYg_faceswap_id01972_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00781\00092_id00476_UgdYVJ6xPYg_faceswap_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00781\00092_id00476_UgdYVJ6xPYg_faceswap_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00830\00143_id00076_Isiq7cA-DNE_faceswap_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00830\00143_id00076_Isiq7cA-DNE_faceswap_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00830\00143_id00076_Isiq7cA-DNE_faceswap_id01544_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00830\00143_id00076_Isiq7cA-DNE_faceswap_id01920_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00944\00135_id01528_SBAS9Kcb8QY_faceswap_id01179_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00987\00160_id02005_7_Egh9mW5y4_faceswap_id01236_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00987\00160_id02005_7_Egh9mW5y4_faceswap_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00987\00160_id02005_7_Egh9mW5y4_faceswap_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00987\00160_id02005_7_Egh9mW5y4_faceswap_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id00987\00160_id02005_7_Egh9mW5y4_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01036\00010_id00701_lW6uzLIOwd0_faceswap_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01036\00010_id00701_lW6uzLIOwd0_faceswap_id01171_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01036\00010_id00701_lW6uzLIOwd0_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01036\00010_id00701_lW6uzLIOwd0_faceswap_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01076\00005_id01207_mt129WTRSII_faceswap_id00391_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01076\00005_id01207_mt129WTRSII_faceswap_id00781_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01076\00005_id01207_mt129WTRSII_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01076\00005_id01207_mt129WTRSII_faceswap_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01076\00005_id01207_mt129WTRSII_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01170\00021_id01933_I5XXxgK7QpE_faceswap_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01170\00021_id01933_I5XXxgK7QpE_faceswap_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01170\00021_id01933_I5XXxgK7QpE_faceswap_id01637_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01170\00021_id01933_I5XXxgK7QpE_faceswap_id01856_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01171\00053_id02494_lObg47hQleE_faceswap_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01171\00053_id02494_lObg47hQleE_faceswap_id00476_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01171\00053_id02494_lObg47hQleE_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01171\00053_id02494_lObg47hQleE_faceswap_id01835_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01171\00053_id02494_lObg47hQleE_faceswap_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01179\00160_id02005_7_Egh9mW5y4_faceswap_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01179\00160_id02005_7_Egh9mW5y4_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01179\00160_id02005_7_Egh9mW5y4_faceswap_id01835_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01179\00160_id02005_7_Egh9mW5y4_faceswap_id01972_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01179\00160_id02005_7_Egh9mW5y4_faceswap_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01207\00320_id00076_Isiq7cA-DNE_faceswap_id00366_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01207\00320_id00076_Isiq7cA-DNE_faceswap_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01207\00320_id00076_Isiq7cA-DNE_faceswap_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01207\00320_id00076_Isiq7cA-DNE_faceswap_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01207\00320_id00076_Isiq7cA-DNE_faceswap_id04727_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01236\00005_id01610_l8zb_iaDJJA_faceswap_id00366_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01236\00005_id01610_l8zb_iaDJJA_faceswap_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01236\00005_id01610_l8zb_iaDJJA_faceswap_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01236\00005_id01610_l8zb_iaDJJA_faceswap_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01236\00005_id01610_l8zb_iaDJJA_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_2_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_2_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_2_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_2_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_2_id01521_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_3_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_3_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_3_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_3_id01933_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01392\00167_3_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01452\00001_id01528_SBAS9Kcb8QY_faceswap_id00173_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01452\00001_id01528_SBAS9Kcb8QY_faceswap_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01452\00001_id01528_SBAS9Kcb8QY_faceswap_id00987_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01452\00001_id01528_SBAS9Kcb8QY_faceswap_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01452\00001_id01528_SBAS9Kcb8QY_faceswap_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01528\00017_id01452_4MqeoSxSy3w_faceswap_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01528\00017_id01452_4MqeoSxSy3w_faceswap_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01528\00017_id01452_4MqeoSxSy3w_faceswap_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01528\00017_id01452_4MqeoSxSy3w_faceswap_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01528\00017_id01452_4MqeoSxSy3w_faceswap_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01544\00044_3_id00166_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01544\00044_3_id00173_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01544\00044_3_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01544\00044_3_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01544\00044_3_id02268_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_0_id00076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_1_id01171_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_1_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_1_id01544_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_1_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_1_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_2_id00987_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_2_id01236_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_2_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_2_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_2_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_3_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_3_id00781_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_3_id01528_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_3_id01920_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01597\00005_3_id02268_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01598\00109_1_id01637_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01598\00109_1_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01598\00109_1_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01598\00109_1_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01598\00109_1_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01610\00090_id01236_7WdumGR5-JM_faceswap_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01610\00090_id01236_7WdumGR5-JM_faceswap_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01610\00090_id01236_7WdumGR5-JM_faceswap_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01610\00090_id01236_7WdumGR5-JM_faceswap_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01610\00090_id01236_7WdumGR5-JM_faceswap_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_1_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_1_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_1_id01920_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_1_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_1_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_2_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_2_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_2_id00944_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_2_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01637\00002_2_id01452_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01691\00045_id01779_HgyHpDEo_jk_faceswap_id00830_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01691\00045_id01779_HgyHpDEo_jk_faceswap_id01236_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01691\00045_id01779_HgyHpDEo_jk_faceswap_id02040_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01691\00045_id01779_HgyHpDEo_jk_faceswap_id02268_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01717\00005_id02005_7_Egh9mW5y4_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01717\00005_id02005_7_Egh9mW5y4_faceswap_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01717\00005_id02005_7_Egh9mW5y4_faceswap_id01691_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01717\00005_id02005_7_Egh9mW5y4_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01717\00005_id02005_7_Egh9mW5y4_faceswap_id04727_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01779\00010_id01691_IVtS5z8Jrrk_faceswap_id00173_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01779\00010_id01691_IVtS5z8Jrrk_faceswap_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01779\00010_id01691_IVtS5z8Jrrk_faceswap_id00701_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01779\00010_id01691_IVtS5z8Jrrk_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01779\00010_id01691_IVtS5z8Jrrk_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01835\00130_id00761_QtTNFhCCgzw_faceswap_id00391_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01835\00130_id00761_QtTNFhCCgzw_faceswap_id00518_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01835\00130_id00761_QtTNFhCCgzw_faceswap_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01835\00130_id00761_QtTNFhCCgzw_faceswap_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01835\00130_id00761_QtTNFhCCgzw_faceswap_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01856\00109_3_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01856\00109_3_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01856\00109_3_id01530_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01856\00109_3_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01856\00109_3_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01920\00099_id00476_UgdYVJ6xPYg_faceswap_id00476_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01920\00099_id00476_UgdYVJ6xPYg_faceswap_id00944_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01920\00099_id00476_UgdYVJ6xPYg_faceswap_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01920\00099_id00476_UgdYVJ6xPYg_faceswap_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01920\00099_id00476_UgdYVJ6xPYg_faceswap_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01933\00028_3_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01933\00028_3_id00518_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01933\00028_3_id00987_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01933\00028_3_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01933\00028_3_id02494_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_2_id01171_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_2_id01179_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_2_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_2_id01597_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_2_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_3_id00366_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_3_id01392_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_3_id01544_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_3_id01779_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id01972\00078_3_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02005\00052_id02342_RJPBPhJB8TA_faceswap_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02005\00052_id02342_RJPBPhJB8TA_faceswap_id00518_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02005\00052_id02342_RJPBPhJB8TA_faceswap_id01207_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02005\00052_id02342_RJPBPhJB8TA_faceswap_id01544_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02005\00052_id02342_RJPBPhJB8TA_faceswap_id01717_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_0_id00761_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_0_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_0_id01835_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_0_id02051_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_0_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_2_id00478_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_2_id01170_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_2_id01452_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_2_id01610_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02296\00019_2_id01920_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02342\00191_id02005_7_Egh9mW5y4_faceswap_id01076_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02342\00191_id02005_7_Egh9mW5y4_faceswap_id01598_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02342\00191_id02005_7_Egh9mW5y4_faceswap_id01835_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02342\00191_id02005_7_Egh9mW5y4_faceswap_id02316_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02342\00191_id02005_7_Egh9mW5y4_faceswap_id02342_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02494\00050_id00475_xQjvXRcnPvw_faceswap_id00475_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02494\00050_id00475_xQjvXRcnPvw_faceswap_id01995_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02494\00050_id00475_xQjvXRcnPvw_faceswap_id02005_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02494\00050_id00475_xQjvXRcnPvw_faceswap_id02296_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\men\id02494\00050_id00475_xQjvXRcnPvw_faceswap_id04727_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00220\00027_id02586_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00220\00027_id03569_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00220\00027_id03658_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00220\00027_id04376_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00220\00027_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00359\00053_id01838_q_lUk55OrL0_faceswap_id00568_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00359\00053_id01838_q_lUk55OrL0_faceswap_id03658_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00359\00053_id01838_q_lUk55OrL0_faceswap_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00359\00053_id01838_q_lUk55OrL0_faceswap_id05106_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00359\00053_id01838_q_lUk55OrL0_faceswap_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00460\00005_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00460\00005_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00460\00005_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00460\00005_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00460\00005_id02808_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00568\00384_id05252_CMxIX3absYM_faceswap_id00577_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00568\00384_id05252_CMxIX3absYM_faceswap_id00707_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00568\00384_id05252_CMxIX3absYM_faceswap_id01661_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00568\00384_id05252_CMxIX3absYM_faceswap_id03747_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00568\00384_id05252_CMxIX3absYM_faceswap_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00577\00010_id01532_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00577\00010_id01907_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00577\00010_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00577\00010_id04376_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00577\00010_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00592\00017_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00592\00017_id01661_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00592\00017_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00592\00017_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00592\00017_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00707\00052_id04820_64ybrA1atlM_faceswap_id02824_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00707\00052_id04820_64ybrA1atlM_faceswap_id04376_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00707\00052_id04820_64ybrA1atlM_faceswap_id04547_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00707\00052_id04820_64ybrA1atlM_faceswap_id04689_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00707\00052_id04820_64ybrA1atlM_faceswap_id04820_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00829\00271_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00829\00271_id03103_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00829\00271_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00829\00271_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00829\00271_id05106_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00832\00078_id00371_t20i0HtPwW0_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00832\00078_id00371_t20i0HtPwW0_faceswap_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00832\00078_id00371_t20i0HtPwW0_faceswap_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00832\00078_id00371_t20i0HtPwW0_faceswap_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id00832\00078_id00371_t20i0HtPwW0_faceswap_id04540_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01178\00028_id05252_CMxIX3absYM_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01178\00028_id05252_CMxIX3absYM_faceswap_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01178\00028_id05252_CMxIX3absYM_faceswap_id02071_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01178\00028_id05252_CMxIX3absYM_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01178\00028_id05252_CMxIX3absYM_faceswap_id04437_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01532\00065_id00220_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01532\00065_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01532\00065_id02508_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01532\00065_id02824_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01532\00065_id05231_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01661\00059_id00577_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01661\00059_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01661\00059_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01661\00059_id02586_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01661\00059_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01783\00015_id02948_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01783\00015_id03569_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01783\00015_id03713_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01783\00015_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01783\00015_id05235_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01838\00126_id05235_ASy8lP3SRtw_faceswap_id00568_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01838\00126_id05235_ASy8lP3SRtw_faceswap_id00829_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01838\00126_id05235_ASy8lP3SRtw_faceswap_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01838\00126_id05235_ASy8lP3SRtw_faceswap_id02071_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01838\00126_id05235_ASy8lP3SRtw_faceswap_id05106_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01907\00148_id05235_ASy8lP3SRtw_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01907\00148_id05235_ASy8lP3SRtw_faceswap_id03656_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01907\00148_id05235_ASy8lP3SRtw_faceswap_id04437_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01907\00148_id05235_ASy8lP3SRtw_faceswap_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id01907\00148_id05235_ASy8lP3SRtw_faceswap_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02301\00092_id00829_aMEvVaUBq2Y_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02301\00092_id00829_aMEvVaUBq2Y_faceswap_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02301\00092_id00829_aMEvVaUBq2Y_faceswap_id02508_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02301\00092_id00829_aMEvVaUBq2Y_faceswap_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02301\00092_id00829_aMEvVaUBq2Y_faceswap_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02508\00083_id03658_8Wtu9VXKqjY_faceswap_id01532_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02508\00083_id03658_8Wtu9VXKqjY_faceswap_id01661_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02508\00083_id03658_8Wtu9VXKqjY_faceswap_id04540_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02508\00083_id03658_8Wtu9VXKqjY_faceswap_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02508\00083_id03658_8Wtu9VXKqjY_faceswap_id05980_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02586\00042_id04939_i4v2cXo9HIQ_faceswap_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02586\00042_id04939_i4v2cXo9HIQ_faceswap_id04245_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02586\00042_id04939_i4v2cXo9HIQ_faceswap_id04374_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02586\00042_id04939_i4v2cXo9HIQ_faceswap_id04820_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02586\00042_id04939_i4v2cXo9HIQ_faceswap_id05106_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02617\00028_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02617\00028_id02838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02617\00028_id03713_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02617\00028_id04689_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02617\00028_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02721\00424_id03658_8Wtu9VXKqjY_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02721\00424_id03658_8Wtu9VXKqjY_faceswap_id02824_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02721\00424_id03658_8Wtu9VXKqjY_faceswap_id02838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02721\00424_id03658_8Wtu9VXKqjY_faceswap_id02948_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02721\00424_id03658_8Wtu9VXKqjY_faceswap_id04820_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02808\00056_id03103_wiCYm3THQPw_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02808\00056_id03103_wiCYm3THQPw_faceswap_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02808\00056_id03103_wiCYm3THQPw_faceswap_id02301_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02808\00056_id03103_wiCYm3THQPw_faceswap_id05235_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02808\00056_id03103_wiCYm3THQPw_faceswap_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02824\00130_id03747_wQOOhZvnrq4_faceswap_id01783_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02824\00130_id03747_wQOOhZvnrq4_faceswap_id02617_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02824\00130_id03747_wQOOhZvnrq4_faceswap_id04245_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02824\00130_id03747_wQOOhZvnrq4_faceswap_id05106_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02824\00130_id03747_wQOOhZvnrq4_faceswap_id05231_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02948\00298_id04820_64ybrA1atlM_faceswap_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02948\00298_id04820_64ybrA1atlM_faceswap_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02948\00298_id04820_64ybrA1atlM_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02948\00298_id04820_64ybrA1atlM_faceswap_id04374_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id02948\00298_id04820_64ybrA1atlM_faceswap_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03569\00065_id00220_WlHLlTQKj8g_faceswap_id00220_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03569\00065_id00220_WlHLlTQKj8g_faceswap_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03569\00065_id00220_WlHLlTQKj8g_faceswap_id02586_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03569\00065_id00220_WlHLlTQKj8g_faceswap_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03569\00065_id00220_WlHLlTQKj8g_faceswap_id05252_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_1_id00359_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_1_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_1_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_1_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_1_id02838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_3_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_3_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_3_id01532_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_3_id02301_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03656\00052_3_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03658\00077_id00371_t20i0HtPwW0_faceswap_id00220_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03658\00077_id00371_t20i0HtPwW0_faceswap_id02301_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03658\00077_id00371_t20i0HtPwW0_faceswap_id03713_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03658\00077_id00371_t20i0HtPwW0_faceswap_id04245_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03658\00077_id00371_t20i0HtPwW0_faceswap_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03713\00249_id02617_4EZjRXC4fLk_faceswap_id00220_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03713\00249_id02617_4EZjRXC4fLk_faceswap_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03713\00249_id02617_4EZjRXC4fLk_faceswap_id01178_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03713\00249_id02617_4EZjRXC4fLk_faceswap_id01532_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03713\00249_id02617_4EZjRXC4fLk_faceswap_id04245_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03747\00273_id02824_glBy_mYcXZw_faceswap_id01661_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03747\00273_id02824_glBy_mYcXZw_faceswap_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03747\00273_id02824_glBy_mYcXZw_faceswap_id04374_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03747\00273_id02824_glBy_mYcXZw_faceswap_id04547_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id03747\00273_id02824_glBy_mYcXZw_faceswap_id05235_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04055\00001_id05252_CMxIX3absYM_faceswap_id00577_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04055\00001_id05252_CMxIX3absYM_faceswap_id03569_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04055\00001_id05252_CMxIX3absYM_faceswap_id04705_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04055\00001_id05252_CMxIX3absYM_faceswap_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04055\00001_id05252_CMxIX3absYM_faceswap_id05980_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04374\00032_id04689_0YqK1ksKjLg_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04374\00032_id04689_0YqK1ksKjLg_faceswap_id01532_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04374\00032_id04689_0YqK1ksKjLg_faceswap_id04689_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04374\00032_id04689_0YqK1ksKjLg_faceswap_id05231_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04376\00181_id04437_2csrqaF55pk_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04376\00181_id04437_2csrqaF55pk_faceswap_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04376\00181_id04437_2csrqaF55pk_faceswap_id00577_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04376\00181_id04437_2csrqaF55pk_faceswap_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04376\00181_id04437_2csrqaF55pk_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04547\00052_2_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04547\00052_2_id02617_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04547\00052_2_id02808_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04547\00052_2_id02824_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04547\00052_2_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04705\00408_id05252_CMxIX3absYM_faceswap_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04705\00408_id05252_CMxIX3absYM_faceswap_id01838_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04705\00408_id05252_CMxIX3absYM_faceswap_id02948_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04705\00408_id05252_CMxIX3absYM_faceswap_id03747_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04705\00408_id05252_CMxIX3absYM_faceswap_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04736\00083_id05235_ASy8lP3SRtw_faceswap_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04736\00083_id05235_ASy8lP3SRtw_faceswap_id01907_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04736\00083_id05235_ASy8lP3SRtw_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04736\00083_id05235_ASy8lP3SRtw_faceswap_id04245_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04820\00015_id02948__ZEDGNWjuFE_faceswap_id00568_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04820\00015_id02948__ZEDGNWjuFE_faceswap_id01783_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04820\00015_id02948__ZEDGNWjuFE_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04820\00015_id02948__ZEDGNWjuFE_faceswap_id04376_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04820\00015_id02948__ZEDGNWjuFE_faceswap_id04689_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04939\00174_id02586_dEYzYDsbAeo_faceswap_id00460_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04939\00174_id02586_dEYzYDsbAeo_faceswap_id01907_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04939\00174_id02586_dEYzYDsbAeo_faceswap_id03747_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04939\00174_id02586_dEYzYDsbAeo_faceswap_id04939_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id04939\00174_id02586_dEYzYDsbAeo_faceswap_id05235_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05106\00078_id04820_64ybrA1atlM_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05106\00078_id04820_64ybrA1atlM_faceswap_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05106\00078_id04820_64ybrA1atlM_faceswap_id01661_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05106\00078_id04820_64ybrA1atlM_faceswap_id04437_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05106\00078_id04820_64ybrA1atlM_faceswap_id05231_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05231\00149_id01178_6XpgYMiKxhc_faceswap_id00371_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05231\00149_id01178_6XpgYMiKxhc_faceswap_id00592_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05231\00149_id01178_6XpgYMiKxhc_faceswap_id01907_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05231\00149_id01178_6XpgYMiKxhc_faceswap_id02301_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05231\00149_id01178_6XpgYMiKxhc_faceswap_id02721_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05235\00052_id01907_LBcRkuRq0uY_faceswap_id01783_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05235\00052_id01907_LBcRkuRq0uY_faceswap_id02808_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05235\00052_id01907_LBcRkuRq0uY_faceswap_id04055_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05235\00052_id01907_LBcRkuRq0uY_faceswap_id04736_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05235\00052_id01907_LBcRkuRq0uY_faceswap_id05251_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05251\00033_id01178_6XpgYMiKxhc_faceswap_id00568_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05251\00033_id01178_6XpgYMiKxhc_faceswap_id03569_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05251\00033_id01178_6XpgYMiKxhc_faceswap_id03658_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05251\00033_id01178_6XpgYMiKxhc_faceswap_id04437_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05252\00052_id01178_6XpgYMiKxhc_faceswap_id00220_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05252\00052_id01178_6XpgYMiKxhc_faceswap_id00832_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05252\00052_id01178_6XpgYMiKxhc_faceswap_id02824_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05252\00052_id01178_6XpgYMiKxhc_faceswap_id04376_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\African\women\id05252\00052_id01178_6XpgYMiKxhc_faceswap_id04437_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00018\00181_id01201_Q8XWfmNiWYA_faceswap_id00243_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00018\00181_id01201_Q8XWfmNiWYA_faceswap_id00777_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00018\00181_id01201_Q8XWfmNiWYA_faceswap_id00945_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00018\00181_id01201_Q8XWfmNiWYA_faceswap_id01239_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00018\00181_id01201_Q8XWfmNiWYA_faceswap_id03678_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00020\00206_id01182_zca-PHR_U40_faceswap_id00018_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00020\00206_id01182_zca-PHR_U40_faceswap_id00049_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00020\00206_id01182_zca-PHR_U40_faceswap_id00696_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00020\00206_id01182_zca-PHR_U40_faceswap_id01048_wavtolip.mp4,fake
+FakeAVCeleb\FakeVideo-FakeAudio\Caucasian (American)\men\id00020\00206_id01182_zca-PHR_U40_faceswap_id01201_wavtolip.mp4,fake

inference.py

@@ -0,0 +1,211 @@
+import os
+import cv2
+import torch
+import argparse
+import numpy as np
+import torch.nn as nn
+from models.TMC import ETMC
+from models import image
+
+#Set random seed for reproducibility.
+torch.manual_seed(42)
+
+
+# Define the audio_args dictionary
+audio_args = {
+    'nb_samp': 64600,
+    'first_conv': 1024,
+    'in_channels': 1,
+    'filts': [20, [20, 20], [20, 128], [128, 128]],
+    'blocks': [2, 4],
+    'nb_fc_node': 1024,
+    'gru_node': 1024,
+    'nb_gru_layer': 3,
+}
+
+
+def get_args(parser):
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--data_dir", type=str, default="datasets/train/fakeavceleb*")
+    parser.add_argument("--LOAD_SIZE", type=int, default=256)
+    parser.add_argument("--FINE_SIZE", type=int, default=224)
+    parser.add_argument("--dropout", type=float, default=0.2)
+    parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
+    parser.add_argument("--hidden", nargs="*", type=int, default=[])
+    parser.add_argument("--hidden_sz", type=int, default=768)
+    parser.add_argument("--img_embed_pool_type", type=str, default="avg", choices=["max", "avg"])
+    parser.add_argument("--img_hidden_sz", type=int, default=1024)
+    parser.add_argument("--include_bn", type=int, default=True)
+    parser.add_argument("--lr", type=float, default=1e-4)
+    parser.add_argument("--lr_factor", type=float, default=0.3)
+    parser.add_argument("--lr_patience", type=int, default=10)
+    parser.add_argument("--max_epochs", type=int, default=500)
+    parser.add_argument("--n_workers", type=int, default=12)
+    parser.add_argument("--name", type=str, default="MMDF")
+    parser.add_argument("--num_image_embeds", type=int, default=1)
+    parser.add_argument("--patience", type=int, default=20)
+    parser.add_argument("--savedir", type=str, default="./savepath/")
+    parser.add_argument("--seed", type=int, default=1)
+    parser.add_argument("--n_classes", type=int, default=2)
+    parser.add_argument("--annealing_epoch", type=int, default=10)
+    parser.add_argument("--device", type=str, default='cpu')
+    parser.add_argument("--pretrained_image_encoder", type=bool, default = False)
+    parser.add_argument("--freeze_image_encoder", type=bool, default = False)
+    parser.add_argument("--pretrained_audio_encoder", type = bool, default=False)
+    parser.add_argument("--freeze_audio_encoder", type = bool, default = False)
+    parser.add_argument("--augment_dataset", type = bool, default = True)
+
+    for key, value in audio_args.items():
+        parser.add_argument(f"--{key}", type=type(value), default=value)
+
+def model_summary(args):
+    '''Prints the model summary.'''
+    model = ETMC(args)
+
+    for name, layer in model.named_modules():
+        print(name, layer)
+
+def load_multimodal_model(args):
+    '''Load multimodal model'''
+    model = ETMC(args)
+    ckpt = torch.load('checkpoints/model_best.pt', map_location = torch.device('cpu'))
+    model.load_state_dict(ckpt,strict = False)
+    model.eval()
+    return model
+
+def load_img_modality_model(args):
+    '''Loads image modality model.'''
+    rgb_encoder = image.ImageEncoder(args)
+    ckpt = torch.load('checkpoints/model_best.pt', map_location = torch.device('cpu'))
+    rgb_encoder.load_state_dict(ckpt,strict = False)
+    rgb_encoder.eval()
+    return rgb_encoder
+
+def load_spec_modality_model(args):
+    spec_encoder = image.RawNet(args)
+    ckpt = torch.load('checkpoints/model_best.pt', map_location = torch.device('cpu'))
+    spec_encoder.load_state_dict(ckpt,strict = False)
+    spec_encoder.eval()
+    return spec_encoder
+
+
+#Load models.
+parser = argparse.ArgumentParser(description="Train Models")
+get_args(parser)
+args, remaining_args = parser.parse_known_args()
+assert remaining_args == [], remaining_args
+
+multimodal = load_multimodal_model(args)
+spec_model = load_spec_modality_model(args)
+img_model = load_img_modality_model(args)
+
+
+def preprocess_img(face):
+    face = face / 255
+    face = cv2.resize(face, (256, 256))
+    face = face.transpose(2, 0, 1) #(W, H, C) -> (C, W, H)
+    face_pt = torch.unsqueeze(torch.Tensor(face), dim = 0) 
+    return face_pt
+
+def preprocess_audio(audio_file):
+    audio_pt = torch.unsqueeze(torch.Tensor(audio_file), dim = 0)
+    return audio_pt
+
+def deepfakes_spec_predict(input_audio):
+    x, _ = input_audio
+    audio = preprocess_audio(x)
+    spec_grads = spec_model.forward(audio)
+    multimodal_grads = multimodal.spec_depth[0].forward(spec_grads)
+
+    out = nn.Softmax()(multimodal_grads)
+    max = torch.argmax(out, dim = -1) #Index of the max value in the tensor.
+    max_value = out[max] #Actual value of the tensor.
+    max_value = np.argmax(out[max].detach().numpy())
+
+    if max_value > 0.5:
+        preds = round(100 - (max_value*100), 3)
+        text2 = f"The audio is REAL."
+
+    else:
+        preds = round(max_value*100, 3)
+        text2 = f"The audio is FAKE."
+
+    return text2
+
+def deepfakes_image_predict(input_image):
+    face = preprocess_img(input_image)
+
+    img_grads = img_model.forward(face)
+    multimodal_grads = multimodal.clf_rgb[0].forward(img_grads)
+
+    out = nn.Softmax()(multimodal_grads)
+    max = torch.argmax(out, dim=-1) #Index of the max value in the tensor.
+    max = max.cpu().detach().numpy()
+    max_value = out[max] #Actual value of the tensor.
+    max_value = np.argmax(out[max].detach().numpy())
+
+    if max_value > 0.5:
+        preds = round(100 - (max_value*100), 3)
+        text2 = f"The image is REAL."
+
+    else:
+        preds = round(max_value*100, 3)
+        text2 = f"The image is FAKE."
+
+    return text2
+
+
+def preprocess_video(input_video, n_frames = 5):
+    v_cap = cv2.VideoCapture(input_video)
+    v_len = int(v_cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+    # Pick 'n_frames' evenly spaced frames to sample
+    if n_frames is None:
+        sample = np.arange(0, v_len)
+    else:
+        sample = np.linspace(0, v_len - 1, n_frames).astype(int)
+
+    #Loop through frames.
+    frames = []
+    for j in range(v_len):
+        success = v_cap.grab()
+        if j in sample:
+            # Load frame
+            success, frame = v_cap.retrieve()
+            if not success:
+                continue
+            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+            frame = preprocess_img(frame)
+            frames.append(frame)
+    v_cap.release()
+    return frames
+
+
+def deepfakes_video_predict(input_video):
+    '''Perform inference on a video.'''
+    video_frames = preprocess_video(input_video)
+
+    real_grads = []
+    fake_grads = []
+
+    for face in video_frames:
+        img_grads = img_model.forward(face)
+        multimodal_grads = multimodal.clf_rgb[0].forward(img_grads)
+
+        out = nn.Softmax()(multimodal_grads)
+        real_grads.append(out.cpu().detach().numpy()[0])
+        print(f"Video out tensor shape is: {out.shape}, {out}")
+
+        fake_grads.append(out.cpu().detach().numpy()[0])
+
+    real_grads_mean = np.mean(real_grads)
+    fake_grads_mean = np.mean(fake_grads)
+
+    if real_grads_mean > fake_grads_mean:
+        res = round(real_grads_mean * 100, 3)
+        text = f"The video is REAL."
+    else:
+        res = round(100 - (real_grads_mean * 100), 3)
+        text = f"The video is FAKE."
+    return text
+

inference_2.py

@@ -0,0 +1,216 @@
+import os
+import cv2
+import onnx
+import torch
+import argparse
+import numpy as np
+import torch.nn as nn
+from models.TMC import ETMC
+from models import image
+
+from onnx2pytorch import ConvertModel
+
+onnx_model = onnx.load('checkpoints/efficientnet.onnx')
+pytorch_model = ConvertModel(onnx_model)
+
+#Set random seed for reproducibility.
+torch.manual_seed(42)
+
+
+# Define the audio_args dictionary
+audio_args = {
+    'nb_samp': 64600,
+    'first_conv': 1024,
+    'in_channels': 1,
+    'filts': [20, [20, 20], [20, 128], [128, 128]],
+    'blocks': [2, 4],
+    'nb_fc_node': 1024,
+    'gru_node': 1024,
+    'nb_gru_layer': 3,
+    'nb_classes': 2
+}
+
+
+def get_args(parser):
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--data_dir", type=str, default="datasets/train/fakeavceleb*")
+    parser.add_argument("--LOAD_SIZE", type=int, default=256)
+    parser.add_argument("--FINE_SIZE", type=int, default=224)
+    parser.add_argument("--dropout", type=float, default=0.2)
+    parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
+    parser.add_argument("--hidden", nargs="*", type=int, default=[])
+    parser.add_argument("--hidden_sz", type=int, default=768)
+    parser.add_argument("--img_embed_pool_type", type=str, default="avg", choices=["max", "avg"])
+    parser.add_argument("--img_hidden_sz", type=int, default=1024)
+    parser.add_argument("--include_bn", type=int, default=True)
+    parser.add_argument("--lr", type=float, default=1e-4)
+    parser.add_argument("--lr_factor", type=float, default=0.3)
+    parser.add_argument("--lr_patience", type=int, default=10)
+    parser.add_argument("--max_epochs", type=int, default=500)
+    parser.add_argument("--n_workers", type=int, default=12)
+    parser.add_argument("--name", type=str, default="MMDF")
+    parser.add_argument("--num_image_embeds", type=int, default=1)
+    parser.add_argument("--patience", type=int, default=20)
+    parser.add_argument("--savedir", type=str, default="./savepath/")
+    parser.add_argument("--seed", type=int, default=1)
+    parser.add_argument("--n_classes", type=int, default=2)
+    parser.add_argument("--annealing_epoch", type=int, default=10)
+    parser.add_argument("--device", type=str, default='cpu')
+    parser.add_argument("--pretrained_image_encoder", type=bool, default = False)
+    parser.add_argument("--freeze_image_encoder", type=bool, default = False)
+    parser.add_argument("--pretrained_audio_encoder", type = bool, default=False)
+    parser.add_argument("--freeze_audio_encoder", type = bool, default = False)
+    parser.add_argument("--augment_dataset", type = bool, default = True)
+
+    for key, value in audio_args.items():
+        parser.add_argument(f"--{key}", type=type(value), default=value)
+
+def model_summary(args):
+    '''Prints the model summary.'''
+    model = ETMC(args)
+
+    for name, layer in model.named_modules():
+        print(name, layer)
+
+def load_multimodal_model(args):
+    '''Load multimodal model'''
+    model = ETMC(args)
+    ckpt = torch.load('checkpoints/model.pth', map_location = torch.device('cpu'))
+    model.load_state_dict(ckpt, strict = True)
+    model.eval()
+    return model
+
+def load_img_modality_model(args):
+    '''Loads image modality model.'''
+    rgb_encoder = pytorch_model
+
+    ckpt = torch.load('checkpoints/model.pth', map_location = torch.device('cpu'))
+    rgb_encoder.load_state_dict(ckpt['rgb_encoder'], strict = True)
+    rgb_encoder.eval()
+    return rgb_encoder
+
+def load_spec_modality_model(args):
+    spec_encoder = image.RawNet(args)
+    ckpt = torch.load('checkpoints/model.pth', map_location = torch.device('cpu'))
+    spec_encoder.load_state_dict(ckpt['spec_encoder'], strict = True)
+    spec_encoder.eval()
+    return spec_encoder
+
+
+#Load models.
+parser = argparse.ArgumentParser(description="Inference models")
+get_args(parser)
+args, remaining_args = parser.parse_known_args()
+assert remaining_args == [], remaining_args
+
+spec_model = load_spec_modality_model(args)
+
+img_model = load_img_modality_model(args)
+
+
+def preprocess_img(face):
+    face = face / 255
+    face = cv2.resize(face, (256, 256))
+    # face = face.transpose(2, 0, 1) #(W, H, C) -> (C, W, H)
+    face_pt = torch.unsqueeze(torch.Tensor(face), dim = 0) 
+    return face_pt
+
+def preprocess_audio(audio_file):
+    audio_pt = torch.unsqueeze(torch.Tensor(audio_file), dim = 0)
+    return audio_pt
+
+def deepfakes_spec_predict(input_audio):
+    x, _ = input_audio
+    audio = preprocess_audio(x)
+    spec_grads = spec_model.forward(audio)
+    spec_grads_inv = np.exp(spec_grads.cpu().detach().numpy().squeeze())
+
+    # multimodal_grads = multimodal.spec_depth[0].forward(spec_grads)
+
+    # out = nn.Softmax()(multimodal_grads)
+    # max = torch.argmax(out, dim = -1) #Index of the max value in the tensor.
+    # max_value = out[max] #Actual value of the tensor.
+    max_value = np.argmax(spec_grads_inv)
+
+    if max_value > 0.5:
+        preds = round(100 - (max_value*100), 3)
+        text2 = f"The audio is REAL."
+
+    else:
+        preds = round(max_value*100, 3)
+        text2 = f"The audio is FAKE."
+
+    return text2
+
+def deepfakes_image_predict(input_image):
+    face = preprocess_img(input_image)
+    print(f"Face shape is: {face.shape}")
+    img_grads = img_model.forward(face)
+    img_grads = img_grads.cpu().detach().numpy()
+    img_grads_np = np.squeeze(img_grads)
+
+    if img_grads_np[0] > 0.5:
+        preds = round(img_grads_np[0] * 100, 3)
+        text2 = f"The image is REAL. \nConfidence score is: {preds}"
+
+    else:
+        preds = round(img_grads_np[1] * 100, 3)
+        text2 = f"The image is FAKE. \nConfidence score is: {preds}"
+
+    return text2
+
+
+def preprocess_video(input_video, n_frames = 3):
+    v_cap = cv2.VideoCapture(input_video)
+    v_len = int(v_cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+    # Pick 'n_frames' evenly spaced frames to sample
+    if n_frames is None:
+        sample = np.arange(0, v_len)
+    else:
+        sample = np.linspace(0, v_len - 1, n_frames).astype(int)
+
+    #Loop through frames.
+    frames = []
+    for j in range(v_len):
+        success = v_cap.grab()
+        if j in sample:
+            # Load frame
+            success, frame = v_cap.retrieve()
+            if not success:
+                continue
+            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+            frame = preprocess_img(frame)
+            frames.append(frame)
+    v_cap.release()
+    return frames
+
+
+def deepfakes_video_predict(input_video):
+    '''Perform inference on a video.'''
+    video_frames = preprocess_video(input_video)
+    real_faces_list = []
+    fake_faces_list = []
+
+    for face in video_frames:
+        # face = preprocess_img(face)
+
+        img_grads = img_model.forward(face)
+        img_grads = img_grads.cpu().detach().numpy()
+        img_grads_np = np.squeeze(img_grads)
+        real_faces_list.append(img_grads_np[0])
+        fake_faces_list.append(img_grads_np[1])
+
+    real_faces_mean = np.mean(real_faces_list)
+    fake_faces_mean = np.mean(fake_faces_list)
+
+    if real_faces_mean > 0.5:
+        preds = round(real_faces_mean * 100, 3)
+        text2 = f"The video is REAL. \nConfidence score is: {preds}%"
+
+    else:
+        preds = round(fake_faces_mean * 100, 3)
+        text2 = f"The video is FAKE. \nConfidence score is: {preds}%"
+
+    return text2
+

main.py

@@ -0,0 +1,247 @@
+import os
+import argparse
+from tqdm import tqdm
+import torch.nn as nn
+import tensorflow as tf
+import torch.optim as optim
+
+from models.TMC import ETMC, ce_loss
+import torchvision.transforms as transforms
+from data.dfdt_dataset import FakeAVCelebDatasetTrain, FakeAVCelebDatasetVal
+
+
+from utils.utils import *
+from utils.logger import create_logger
+from sklearn.metrics import accuracy_score
+from torch.utils.tensorboard import SummaryWriter
+
+# Define the audio_args dictionary
+audio_args = {
+    'nb_samp': 64600,
+    'first_conv': 1024,
+    'in_channels': 1,
+    'filts': [20, [20, 20], [20, 128], [128, 128]],
+    'blocks': [2, 4],
+    'nb_fc_node': 1024,
+    'gru_node': 1024,
+    'nb_gru_layer': 3,
+}
+
+
+def get_args(parser):
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--data_dir", type=str, default="datasets/train/fakeavceleb*")
+    parser.add_argument("--LOAD_SIZE", type=int, default=256)
+    parser.add_argument("--FINE_SIZE", type=int, default=224)
+    parser.add_argument("--dropout", type=float, default=0.2)
+    parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
+    parser.add_argument("--hidden", nargs="*", type=int, default=[])
+    parser.add_argument("--hidden_sz", type=int, default=768)
+    parser.add_argument("--img_embed_pool_type", type=str, default="avg", choices=["max", "avg"])
+    parser.add_argument("--img_hidden_sz", type=int, default=1024)
+    parser.add_argument("--include_bn", type=int, default=True)
+    parser.add_argument("--lr", type=float, default=1e-4)
+    parser.add_argument("--lr_factor", type=float, default=0.3)
+    parser.add_argument("--lr_patience", type=int, default=10)
+    parser.add_argument("--max_epochs", type=int, default=500)
+    parser.add_argument("--n_workers", type=int, default=12)
+    parser.add_argument("--name", type=str, default="MMDF")
+    parser.add_argument("--num_image_embeds", type=int, default=1)
+    parser.add_argument("--patience", type=int, default=20)
+    parser.add_argument("--savedir", type=str, default="./savepath/")
+    parser.add_argument("--seed", type=int, default=1)
+    parser.add_argument("--n_classes", type=int, default=2)
+    parser.add_argument("--annealing_epoch", type=int, default=10)
+    parser.add_argument("--device", type=str, default='cpu')
+    parser.add_argument("--pretrained_image_encoder", type=bool, default = False)
+    parser.add_argument("--freeze_image_encoder", type=bool, default = True)
+    parser.add_argument("--pretrained_audio_encoder", type = bool, default=False)
+    parser.add_argument("--freeze_audio_encoder", type = bool, default = True)
+    parser.add_argument("--augment_dataset", type = bool, default = True)
+
+    for key, value in audio_args.items():
+        parser.add_argument(f"--{key}", type=type(value), default=value)
+
+def get_optimizer(model, args):
+    optimizer = optim.Adam(model.parameters(), lr=args.lr, weight_decay=1e-5)
+    return optimizer
+
+
+def get_scheduler(optimizer, args):
+    return optim.lr_scheduler.ReduceLROnPlateau(
+        optimizer, "max", patience=args.lr_patience, verbose=True, factor=args.lr_factor
+    )
+
+def model_forward(i_epoch, model, args, ce_loss, batch):
+    rgb, spec, tgt = batch['video_reshaped'], batch['spectrogram'], batch['label_map']
+    rgb_pt = torch.Tensor(rgb.numpy())
+    spec = spec.numpy()
+    spec_pt = torch.Tensor(spec)
+    tgt_pt = torch.Tensor(tgt.numpy())
+
+    if torch.cuda.is_available():
+        rgb_pt, spec_pt, tgt_pt = rgb_pt.cuda(), spec_pt.cuda(), tgt_pt.cuda()
+        
+    # depth_alpha, rgb_alpha, depth_rgb_alpha = model(rgb_pt, spec_pt)
+
+    # loss = ce_loss(tgt_pt, depth_alpha, args.n_classes, i_epoch, args.annealing_epoch) + \
+    #        ce_loss(tgt_pt, rgb_alpha, args.n_classes, i_epoch, args.annealing_epoch) + \
+    #        ce_loss(tgt_pt, depth_rgb_alpha, args.n_classes, i_epoch, args.annealing_epoch)
+    # return loss, depth_alpha, rgb_alpha, depth_rgb_alpha, tgt_pt
+
+    depth_alpha, rgb_alpha, pseudo_alpha, depth_rgb_alpha = model(rgb_pt, spec_pt)
+
+    loss = ce_loss(tgt_pt, depth_alpha, args.n_classes, i_epoch, args.annealing_epoch) + \
+           ce_loss(tgt_pt, rgb_alpha, args.n_classes, i_epoch, args.annealing_epoch) + \
+           ce_loss(tgt_pt, pseudo_alpha, args.n_classes, i_epoch, args.annealing_epoch) + \
+           ce_loss(tgt_pt, depth_rgb_alpha, args.n_classes, i_epoch, args.annealing_epoch)
+    return loss, depth_alpha, rgb_alpha, depth_rgb_alpha, tgt_pt
+
+
+
+def model_eval(i_epoch, data, model, args, criterion):
+    model.eval()
+    with torch.no_grad():
+        losses, depth_preds, rgb_preds, depthrgb_preds, tgts = [], [], [], [], []
+        for batch in tqdm(data):
+            loss, depth_alpha, rgb_alpha, depth_rgb_alpha, tgt = model_forward(i_epoch, model, args, criterion, batch)
+            losses.append(loss.item())
+
+            depth_pred = depth_alpha.argmax(dim=1).cpu().detach().numpy()
+            rgb_pred = rgb_alpha.argmax(dim=1).cpu().detach().numpy()
+            depth_rgb_pred = depth_rgb_alpha.argmax(dim=1).cpu().detach().numpy()
+
+            depth_preds.append(depth_pred)
+            rgb_preds.append(rgb_pred)
+            depthrgb_preds.append(depth_rgb_pred)
+            tgt = tgt.cpu().detach().numpy()
+            tgts.append(tgt)
+
+    metrics = {"loss": np.mean(losses)}
+    print(f"Mean loss is: {metrics['loss']}")
+
+    tgts = [l for sl in tgts for l in sl]
+    depth_preds = [l for sl in depth_preds for l in sl]
+    rgb_preds = [l for sl in rgb_preds for l in sl]
+    depthrgb_preds = [l for sl in depthrgb_preds for l in sl]
+    metrics["spec_acc"] = accuracy_score(tgts, depth_preds)
+    metrics["rgb_acc"] = accuracy_score(tgts, rgb_preds)
+    metrics["specrgb_acc"] = accuracy_score(tgts, depthrgb_preds)
+    return metrics
+
+def write_weight_histograms(writer, step, model):
+    for idx, item in enumerate(model.named_parameters()):
+        name = item[0]
+        weights = item[1].data
+        if weights.size(dim = 0) > 2:
+            try:
+                writer.add_histogram(name, weights, idx)
+            except ValueError as e:
+                continue
+
+writer = SummaryWriter()
+
+def train(args):
+    set_seed(args.seed)
+    args.savedir = os.path.join(args.savedir, args.name)
+    os.makedirs(args.savedir, exist_ok=True)
+
+    train_ds = FakeAVCelebDatasetTrain(args)
+    train_ds = train_ds.load_features_from_tfrec()
+
+    val_ds = FakeAVCelebDatasetVal(args)
+    val_ds = val_ds.load_features_from_tfrec()
+    
+    model = ETMC(args)
+    optimizer = get_optimizer(model, args)
+    scheduler = get_scheduler(optimizer, args)
+    logger = create_logger("%s/logfile.log" % args.savedir, args)
+    if torch.cuda.is_available():
+        model.cuda()
+
+    torch.save(args, os.path.join(args.savedir, "checkpoint.pt"))
+    start_epoch, global_step, n_no_improve, best_metric = 0, 0, 0, -np.inf
+
+    for i_epoch in range(start_epoch, args.max_epochs):
+        train_losses = []
+        model.train()
+        optimizer.zero_grad()
+
+        for index, batch in tqdm(enumerate(train_ds)):
+            loss, depth_out, rgb_out, depthrgb, tgt = model_forward(i_epoch, model, args, ce_loss, batch)
+            if args.gradient_accumulation_steps > 1:
+                 loss = loss / args.gradient_accumulation_steps
+
+            train_losses.append(loss.item())
+            loss.backward()
+            global_step += 1
+            if global_step % args.gradient_accumulation_steps == 0:
+                optimizer.step()
+                optimizer.zero_grad()
+
+        #Write weight histograms to Tensorboard.
+        write_weight_histograms(writer, i_epoch, model)
+
+        model.eval()
+        metrics = model_eval(
+            np.inf, val_ds, model, args, ce_loss
+        )
+        logger.info("Train Loss: {:.4f}".format(np.mean(train_losses)))
+        log_metrics("val", metrics, logger)
+        logger.info(
+            "{}: Loss: {:.5f} | spec_acc: {:.5f}, rgb_acc: {:.5f}, depth rgb acc: {:.5f}".format(
+                "val", metrics["loss"], metrics["spec_acc"], metrics["rgb_acc"], metrics["specrgb_acc"]
+            )
+        )
+        tuning_metric = metrics["specrgb_acc"]
+
+        scheduler.step(tuning_metric)
+        is_improvement = tuning_metric > best_metric
+        if is_improvement:
+            best_metric = tuning_metric
+            n_no_improve = 0
+        else:
+            n_no_improve += 1
+
+        save_checkpoint(
+            {
+                "epoch": i_epoch + 1,
+                "optimizer": optimizer.state_dict(),
+                "scheduler": scheduler.state_dict(),
+                "n_no_improve": n_no_improve,
+                "best_metric": best_metric,
+            },
+            is_improvement,
+            args.savedir,
+        )
+
+        if n_no_improve >= args.patience:
+            logger.info("No improvement. Breaking out of loop.")
+            break
+    writer.close()
+    # load_checkpoint(model, os.path.join(args.savedir, "model_best.pt"))
+    model.eval()
+    test_metrics = model_eval(
+        np.inf, val_ds, model, args, ce_loss
+    )
+    logger.info(
+        "{}: Loss: {:.5f} | spec_acc: {:.5f}, rgb_acc: {:.5f}, depth rgb acc: {:.5f}".format(
+            "Test", test_metrics["loss"], test_metrics["spec_acc"], test_metrics["rgb_acc"],
+            test_metrics["depthrgb_acc"]
+        )
+    )
+    log_metrics(f"Test", test_metrics, logger)
+
+
+def cli_main():
+    parser = argparse.ArgumentParser(description="Train Models")
+    get_args(parser)
+    args, remaining_args = parser.parse_known_args()
+    assert remaining_args == [], remaining_args
+    train(args)
+
+
+if __name__ == "__main__":
+    import warnings
+    warnings.filterwarnings("ignore")
+    cli_main()

models/TMC.py

@@ -0,0 +1,156 @@
+import torch
+import torch.nn as nn
+from models import image
+import torch.nn.functional as F
+
+
+# loss function
+def KL(alpha, c):
+    if torch.cuda.is_available():
+        beta = torch.ones((1, c)).cuda()
+    else:
+        beta = torch.ones((1, c))
+    S_alpha = torch.sum(alpha, dim=1, keepdim=True)
+    S_beta = torch.sum(beta, dim=1, keepdim=True)
+    lnB = torch.lgamma(S_alpha) - torch.sum(torch.lgamma(alpha), dim=1, keepdim=True)
+    lnB_uni = torch.sum(torch.lgamma(beta), dim=1, keepdim=True) - torch.lgamma(S_beta)
+    dg0 = torch.digamma(S_alpha)
+    dg1 = torch.digamma(alpha)
+    kl = torch.sum((alpha - beta) * (dg1 - dg0), dim=1, keepdim=True) + lnB + lnB_uni
+    return kl
+
+def ce_loss(p, alpha, c, global_step, annealing_step):
+    S = torch.sum(alpha, dim=1, keepdim=True)
+    E = alpha - 1
+    label = p
+    A = torch.sum(label * (torch.digamma(S) - torch.digamma(alpha)), dim=1, keepdim=True)
+
+    annealing_coef = min(1, global_step / annealing_step)
+    alp = E * (1 - label) + 1
+    B = annealing_coef * KL(alp, c)
+    return torch.mean((A + B))
+
+
+class TMC(nn.Module):
+    def __init__(self, args):
+        super(TMC, self).__init__()
+        self.args = args
+        self.rgbenc = image.ImageEncoder(args)
+        self.specenc = image.RawNet(args)
+        
+        spec_last_size = args.img_hidden_sz * 1
+        rgb_last_size = args.img_hidden_sz * args.num_image_embeds
+        self.spec_depth = nn.ModuleList()
+        self.clf_rgb = nn.ModuleList()
+
+        for hidden in args.hidden:
+            self.spec_depth.append(nn.Linear(spec_last_size, hidden))
+            self.spec_depth.append(nn.ReLU())
+            self.spec_depth.append(nn.Dropout(args.dropout))
+            spec_last_size = hidden
+        self.spec_depth.append(nn.Linear(spec_last_size, args.n_classes))
+
+        for hidden in args.hidden:
+            self.clf_rgb.append(nn.Linear(rgb_last_size, hidden))
+            self.clf_rgb.append(nn.ReLU())
+            self.clf_rgb.append(nn.Dropout(args.dropout))
+            rgb_last_size = hidden
+        self.clf_rgb.append(nn.Linear(rgb_last_size, args.n_classes))
+
+    def DS_Combin_two(self, alpha1, alpha2):
+        # Calculate the merger of two DS evidences
+        alpha = dict()
+        alpha[0], alpha[1] = alpha1, alpha2
+        b, S, E, u = dict(), dict(), dict(), dict()
+        for v in range(2):
+            S[v] = torch.sum(alpha[v], dim=1, keepdim=True)
+            E[v] = alpha[v] - 1
+            b[v] = E[v] / (S[v].expand(E[v].shape))
+            u[v] = self.args.n_classes / S[v]
+
+        # b^0 @ b^(0+1)
+        bb = torch.bmm(b[0].view(-1, self.args.n_classes, 1), b[1].view(-1, 1, self.args.n_classes))
+        # b^0 * u^1
+        uv1_expand = u[1].expand(b[0].shape)
+        bu = torch.mul(b[0], uv1_expand)
+        # b^1 * u^0
+        uv_expand = u[0].expand(b[0].shape)
+        ub = torch.mul(b[1], uv_expand)
+        # calculate K
+        bb_sum = torch.sum(bb, dim=(1, 2), out=None)
+        bb_diag = torch.diagonal(bb, dim1=-2, dim2=-1).sum(-1)
+        # bb_diag1 = torch.diag(torch.mm(b[v], torch.transpose(b[v+1], 0, 1)))
+        K = bb_sum - bb_diag
+
+        # calculate b^a
+        b_a = (torch.mul(b[0], b[1]) + bu + ub) / ((1 - K).view(-1, 1).expand(b[0].shape))
+        # calculate u^a
+        u_a = torch.mul(u[0], u[1]) / ((1 - K).view(-1, 1).expand(u[0].shape))
+        # test = torch.sum(b_a, dim = 1, keepdim = True) + u_a #Verify programming errors
+
+        # calculate new S
+        S_a = self.args.n_classes / u_a
+        # calculate new e_k
+        e_a = torch.mul(b_a, S_a.expand(b_a.shape))
+        alpha_a = e_a + 1
+        return alpha_a
+
+    def forward(self, rgb, spec):
+        spec = self.specenc(spec)
+        spec = torch.flatten(spec, start_dim=1)
+
+        rgb = self.rgbenc(rgb)
+        rgb = torch.flatten(rgb, start_dim=1)
+
+        spec_out = spec
+
+        for layer in self.spec_depth:
+            spec_out = layer(spec_out)
+
+        rgb_out = rgb
+
+        for layer in self.clf_rgb:
+            rgb_out = layer(rgb_out)
+
+        spec_evidence, rgb_evidence = F.softplus(spec_out), F.softplus(rgb_out)
+        spec_alpha, rgb_alpha = spec_evidence+1, rgb_evidence+1
+        spec_rgb_alpha = self.DS_Combin_two(spec_alpha, rgb_alpha)
+        return spec_alpha, rgb_alpha, spec_rgb_alpha
+
+
+class ETMC(TMC):
+    def __init__(self, args):
+        super(ETMC, self).__init__(args)
+        last_size = args.img_hidden_sz * args.num_image_embeds + args.img_hidden_sz * args.num_image_embeds
+        self.clf = nn.ModuleList()
+        for hidden in args.hidden:
+            self.clf.append(nn.Linear(last_size, hidden))
+            self.clf.append(nn.ReLU())
+            self.clf.append(nn.Dropout(args.dropout))
+            last_size = hidden
+        self.clf.append(nn.Linear(last_size, args.n_classes))
+
+    def forward(self, rgb, spec):
+        spec = self.specenc(spec)
+        spec = torch.flatten(spec, start_dim=1)
+
+        rgb = self.rgbenc(rgb)
+        rgb = torch.flatten(rgb, start_dim=1)
+
+        spec_out = spec
+        for layer in self.spec_depth:
+            spec_out = layer(spec_out)
+
+        rgb_out = rgb
+        for layer in self.clf_rgb:
+            rgb_out = layer(rgb_out)
+
+        pseudo_out = torch.cat([rgb, spec], -1)
+        for layer in self.clf:
+            pseudo_out = layer(pseudo_out)
+
+        depth_evidence, rgb_evidence, pseudo_evidence = F.softplus(spec_out), F.softplus(rgb_out), F.softplus(pseudo_out)
+        depth_alpha, rgb_alpha, pseudo_alpha = depth_evidence+1, rgb_evidence+1, pseudo_evidence+1
+        depth_rgb_alpha = self.DS_Combin_two(self.DS_Combin_two(depth_alpha, rgb_alpha), pseudo_alpha)
+        return depth_alpha, rgb_alpha, pseudo_alpha, depth_rgb_alpha
+

models/classifiers.py

@@ -0,0 +1,172 @@
+from functools import partial
+
+import numpy as np
+import torch
+from timm.models.efficientnet import tf_efficientnet_b4_ns, tf_efficientnet_b3_ns, \
+    tf_efficientnet_b5_ns, tf_efficientnet_b2_ns, tf_efficientnet_b6_ns, tf_efficientnet_b7_ns
+from torch import nn
+from torch.nn.modules.dropout import Dropout
+from torch.nn.modules.linear import Linear
+from torch.nn.modules.pooling import AdaptiveAvgPool2d
+
+encoder_params = {
+    "tf_efficientnet_b3_ns": {
+        "features": 1536,
+        "init_op": partial(tf_efficientnet_b3_ns, pretrained=True, drop_path_rate=0.2)
+    },
+    "tf_efficientnet_b2_ns": {
+        "features": 1408,
+        "init_op": partial(tf_efficientnet_b2_ns, pretrained=False, drop_path_rate=0.2)
+    },
+    "tf_efficientnet_b4_ns": {
+        "features": 1792,
+        "init_op": partial(tf_efficientnet_b4_ns, pretrained=True, drop_path_rate=0.5)
+    },
+    "tf_efficientnet_b5_ns": {
+        "features": 2048,
+        "init_op": partial(tf_efficientnet_b5_ns, pretrained=True, drop_path_rate=0.2)
+    },
+    "tf_efficientnet_b4_ns_03d": {
+        "features": 1792,
+        "init_op": partial(tf_efficientnet_b4_ns, pretrained=True, drop_path_rate=0.3)
+    },
+    "tf_efficientnet_b5_ns_03d": {
+        "features": 2048,
+        "init_op": partial(tf_efficientnet_b5_ns, pretrained=True, drop_path_rate=0.3)
+    },
+    "tf_efficientnet_b5_ns_04d": {
+        "features": 2048,
+        "init_op": partial(tf_efficientnet_b5_ns, pretrained=True, drop_path_rate=0.4)
+    },
+    "tf_efficientnet_b6_ns": {
+        "features": 2304,
+        "init_op": partial(tf_efficientnet_b6_ns, pretrained=True, drop_path_rate=0.2)
+    },
+    "tf_efficientnet_b7_ns": {
+        "features": 2560,
+        "init_op": partial(tf_efficientnet_b7_ns, pretrained=False, drop_path_rate=0.2)
+    },
+    "tf_efficientnet_b6_ns_04d": {
+        "features": 2304,
+        "init_op": partial(tf_efficientnet_b6_ns, pretrained=True, drop_path_rate=0.4)
+    },
+}
+
+
+def setup_srm_weights(input_channels: int = 3) -> torch.Tensor:
+    """Creates the SRM kernels for noise analysis."""
+    # note: values taken from Zhou et al., "Learning Rich Features for Image Manipulation Detection", CVPR2018
+    srm_kernel = torch.from_numpy(np.array([
+        [  # srm 1/2 horiz
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+            [0., 1., -2., 1., 0.],  # noqa: E241,E201
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+        ], [  # srm 1/4
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+            [0., -1., 2., -1., 0.],  # noqa: E241,E201
+            [0., 2., -4., 2., 0.],  # noqa: E241,E201
+            [0., -1., 2., -1., 0.],  # noqa: E241,E201
+            [0., 0., 0., 0., 0.],  # noqa: E241,E201
+        ], [  # srm 1/12
+            [-1., 2., -2., 2., -1.],  # noqa: E241,E201
+            [2., -6., 8., -6., 2.],  # noqa: E241,E201
+            [-2., 8., -12., 8., -2.],  # noqa: E241,E201
+            [2., -6., 8., -6., 2.],  # noqa: E241,E201
+            [-1., 2., -2., 2., -1.],  # noqa: E241,E201
+        ]
+    ])).float()
+    srm_kernel[0] /= 2
+    srm_kernel[1] /= 4
+    srm_kernel[2] /= 12
+    return srm_kernel.view(3, 1, 5, 5).repeat(1, input_channels, 1, 1)
+
+
+def setup_srm_layer(input_channels: int = 3) -> torch.nn.Module:
+    """Creates a SRM convolution layer for noise analysis."""
+    weights = setup_srm_weights(input_channels)
+    conv = torch.nn.Conv2d(input_channels, out_channels=3, kernel_size=5, stride=1, padding=2, bias=False)
+    with torch.no_grad():
+        conv.weight = torch.nn.Parameter(weights, requires_grad=False)
+    return conv
+
+
+class DeepFakeClassifierSRM(nn.Module):
+    def __init__(self, encoder, dropout_rate=0.5) -> None:
+        super().__init__()
+        self.encoder = encoder_params[encoder]["init_op"]()
+        self.avg_pool = AdaptiveAvgPool2d((1, 1))
+        self.srm_conv = setup_srm_layer(3)
+        self.dropout = Dropout(dropout_rate)
+        self.fc = Linear(encoder_params[encoder]["features"], 1)
+
+    def forward(self, x):
+        noise = self.srm_conv(x)
+        x = self.encoder.forward_features(noise)
+        x = self.avg_pool(x).flatten(1)
+        x = self.dropout(x)
+        x = self.fc(x)
+        return x
+
+
+class GlobalWeightedAvgPool2d(nn.Module):
+    """
+    Global Weighted Average Pooling from paper "Global Weighted Average
+    Pooling Bridges Pixel-level Localization and Image-level Classification"
+    """
+
+    def __init__(self, features: int, flatten=False):
+        super().__init__()
+        self.conv = nn.Conv2d(features, 1, kernel_size=1, bias=True)
+        self.flatten = flatten
+
+    def fscore(self, x):
+        m = self.conv(x)
+        m = m.sigmoid().exp()
+        return m
+
+    def norm(self, x: torch.Tensor):
+        return x / x.sum(dim=[2, 3], keepdim=True)
+
+    def forward(self, x):
+        input_x = x
+        x = self.fscore(x)
+        x = self.norm(x)
+        x = x * input_x
+        x = x.sum(dim=[2, 3], keepdim=not self.flatten)
+        return x
+
+
+class DeepFakeClassifier(nn.Module):
+    def __init__(self, encoder, dropout_rate=0.0) -> None:
+        super().__init__()
+        self.encoder = encoder_params[encoder]["init_op"]()
+        self.avg_pool = AdaptiveAvgPool2d((1, 1))
+        self.dropout = Dropout(dropout_rate)
+        self.fc = Linear(encoder_params[encoder]["features"], 1)
+
+    def forward(self, x):
+        x = self.encoder.forward_features(x)
+        x = self.avg_pool(x).flatten(1)
+        x = self.dropout(x)
+        x = self.fc(x)
+        return x
+
+
+
+
+class DeepFakeClassifierGWAP(nn.Module):
+    def __init__(self, encoder, dropout_rate=0.5) -> None:
+        super().__init__()
+        self.encoder = encoder_params[encoder]["init_op"]()
+        self.avg_pool = GlobalWeightedAvgPool2d(encoder_params[encoder]["features"])
+        self.dropout = Dropout(dropout_rate)
+        self.fc = Linear(encoder_params[encoder]["features"], 1)
+
+    def forward(self, x):
+        x = self.encoder.forward_features(x)
+        x = self.avg_pool(x).flatten(1)
+        x = self.dropout(x)
+        x = self.fc(x)
+        return x

models/image.py

@@ -0,0 +1,195 @@
+import re
+import os
+import wget
+import torch
+import torchvision
+import torch.nn as nn
+import torch.nn.functional as F
+from models.rawnet import SincConv, Residual_block
+from models.classifiers import DeepFakeClassifier
+
+class ImageEncoder(nn.Module):
+    def __init__(self, args):
+        super(ImageEncoder, self).__init__()
+        self.device = args.device
+        self.args = args
+        self.flatten = nn.Flatten()
+        self.sigmoid = nn.Sigmoid()
+        # self.fc = nn.Linear(in_features=2560, out_features = 2)
+        self.pretrained_image_encoder = args.pretrained_image_encoder
+        self.freeze_image_encoder = args.freeze_image_encoder
+
+        if self.pretrained_image_encoder == False:
+            self.model = DeepFakeClassifier(encoder = "tf_efficientnet_b7_ns").to(self.device)
+
+        else:
+            self.pretrained_ckpt = torch.load('pretrained\\final_999_DeepFakeClassifier_tf_efficientnet_b7_ns_0_23', map_location = torch.device(self.args.device))
+            self.state_dict = self.pretrained_ckpt.get("state_dict", self.pretrained_ckpt)
+
+            self.model = DeepFakeClassifier(encoder = "tf_efficientnet_b7_ns").to(self.device)
+            print("Loading pretrained image encoder...")
+            self.model.load_state_dict({re.sub("^module.", "", k): v for k, v in self.state_dict.items()}, strict=True)
+            print("Loaded pretrained image encoder.")
+
+        if self.freeze_image_encoder == True:
+            for idx, param in self.model.named_parameters():
+                param.requires_grad = False
+
+        # self.model.fc = nn.Identity()
+
+    def forward(self, x):
+        x = self.model(x)
+        out = self.sigmoid(x)
+        # x = self.flatten(x)
+        # out = self.fc(x)
+        return out
+
+
+class RawNet(nn.Module):
+    def __init__(self, args):
+        super(RawNet, self).__init__()
+        
+        self.device=args.device
+        self.filts = [20, [20, 20], [20, 128], [128, 128]]
+
+        self.Sinc_conv=SincConv(device=self.device,
+			out_channels = self.filts[0],
+			kernel_size = 1024,
+            in_channels = args.in_channels)
+        
+        self.first_bn = nn.BatchNorm1d(num_features = self.filts[0])
+        self.selu = nn.SELU(inplace=True)
+        self.block0 = nn.Sequential(Residual_block(nb_filts = self.filts[1], first = True))
+        self.block1 = nn.Sequential(Residual_block(nb_filts = self.filts[1]))
+        self.block2 = nn.Sequential(Residual_block(nb_filts = self.filts[2]))
+        self.filts[2][0] = self.filts[2][1]
+        self.block3 = nn.Sequential(Residual_block(nb_filts = self.filts[2]))
+        self.block4 = nn.Sequential(Residual_block(nb_filts = self.filts[2]))
+        self.block5 = nn.Sequential(Residual_block(nb_filts = self.filts[2]))
+        self.avgpool = nn.AdaptiveAvgPool1d(1)
+
+        self.fc_attention0 = self._make_attention_fc(in_features = self.filts[1][-1],
+            l_out_features = self.filts[1][-1])
+        self.fc_attention1 = self._make_attention_fc(in_features = self.filts[1][-1],
+            l_out_features = self.filts[1][-1])
+        self.fc_attention2 = self._make_attention_fc(in_features = self.filts[2][-1],
+            l_out_features = self.filts[2][-1])
+        self.fc_attention3 = self._make_attention_fc(in_features = self.filts[2][-1],
+            l_out_features = self.filts[2][-1])
+        self.fc_attention4 = self._make_attention_fc(in_features = self.filts[2][-1],
+            l_out_features = self.filts[2][-1])
+        self.fc_attention5 = self._make_attention_fc(in_features = self.filts[2][-1],
+            l_out_features = self.filts[2][-1])
+
+        self.bn_before_gru = nn.BatchNorm1d(num_features = self.filts[2][-1])
+        self.gru = nn.GRU(input_size = self.filts[2][-1],
+			hidden_size = args.gru_node,
+			num_layers = args.nb_gru_layer,
+			batch_first = True)
+        
+        self.fc1_gru = nn.Linear(in_features = args.gru_node,
+			out_features = args.nb_fc_node)
+       
+        self.fc2_gru = nn.Linear(in_features = args.nb_fc_node,
+			out_features = args.nb_classes ,bias=True)
+       
+        self.sig = nn.Sigmoid()
+        self.logsoftmax = nn.LogSoftmax(dim=1)
+        self.pretrained_audio_encoder = args.pretrained_audio_encoder
+        self.freeze_audio_encoder = args.freeze_audio_encoder
+
+        if self.pretrained_audio_encoder == True:
+            print("Loading pretrained audio encoder")
+            ckpt = torch.load('pretrained\\RawNet.pth', map_location = torch.device(self.device))
+            print("Loaded pretrained audio encoder")
+            self.load_state_dict(ckpt, strict = True)
+        
+        if self.freeze_audio_encoder:
+                for param in self.parameters():
+                    param.requires_grad = False
+
+
+    def forward(self, x, y = None):
+
+        nb_samp = x.shape[0]
+        len_seq = x.shape[1]
+        x=x.view(nb_samp,1,len_seq)
+        
+        x = self.Sinc_conv(x)    
+        x = F.max_pool1d(torch.abs(x), 3)
+        x = self.first_bn(x)
+        x =  self.selu(x)
+        
+        x0 = self.block0(x)
+        y0 = self.avgpool(x0).view(x0.size(0), -1) # torch.Size([batch, filter])
+        y0 = self.fc_attention0(y0)
+        y0 = self.sig(y0).view(y0.size(0), y0.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x0 * y0 + y0  # (batch, filter, time) x (batch, filter, 1)
+        
+
+        x1 = self.block1(x)
+        y1 = self.avgpool(x1).view(x1.size(0), -1) # torch.Size([batch, filter])
+        y1 = self.fc_attention1(y1)
+        y1 = self.sig(y1).view(y1.size(0), y1.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x1 * y1 + y1 # (batch, filter, time) x (batch, filter, 1)
+
+        x2 = self.block2(x)
+        y2 = self.avgpool(x2).view(x2.size(0), -1) # torch.Size([batch, filter])
+        y2 = self.fc_attention2(y2)
+        y2 = self.sig(y2).view(y2.size(0), y2.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x2 * y2 + y2 # (batch, filter, time) x (batch, filter, 1)
+
+        x3 = self.block3(x)
+        y3 = self.avgpool(x3).view(x3.size(0), -1) # torch.Size([batch, filter])
+        y3 = self.fc_attention3(y3)
+        y3 = self.sig(y3).view(y3.size(0), y3.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x3 * y3 + y3 # (batch, filter, time) x (batch, filter, 1)
+
+        x4 = self.block4(x)
+        y4 = self.avgpool(x4).view(x4.size(0), -1) # torch.Size([batch, filter])
+        y4 = self.fc_attention4(y4)
+        y4 = self.sig(y4).view(y4.size(0), y4.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x4 * y4 + y4 # (batch, filter, time) x (batch, filter, 1)
+
+        x5 = self.block5(x)
+        y5 = self.avgpool(x5).view(x5.size(0), -1) # torch.Size([batch, filter])
+        y5 = self.fc_attention5(y5)
+        y5 = self.sig(y5).view(y5.size(0), y5.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x5 * y5 + y5 # (batch, filter, time) x (batch, filter, 1)
+
+        x = self.bn_before_gru(x)
+        x = self.selu(x)
+        x = x.permute(0, 2, 1)     #(batch, filt, time) >> (batch, time, filt)
+        self.gru.flatten_parameters()
+        x, _ = self.gru(x)
+        x = x[:,-1,:]
+        x = self.fc1_gru(x)
+        x = self.fc2_gru(x)
+        output=self.logsoftmax(x)
+      
+        return output
+        
+        
+
+    def _make_attention_fc(self, in_features, l_out_features):
+
+        l_fc = []
+        
+        l_fc.append(nn.Linear(in_features = in_features,
+			        out_features = l_out_features))
+
+        
+
+        return nn.Sequential(*l_fc)
+
+
+    def _make_layer(self, nb_blocks, nb_filts, first = False):
+        layers = []
+        #def __init__(self, nb_filts, first = False):
+        for i in range(nb_blocks):
+            first = first if i == 0 else False
+            layers.append(Residual_block(nb_filts = nb_filts,
+				first = first))
+            if i == 0: nb_filts[0] = nb_filts[1]
+            
+        return nn.Sequential(*layers)

models/rawnet.py

@@ -0,0 +1,360 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch import Tensor
+import numpy as np
+from torch.utils import data
+from collections import OrderedDict
+from torch.nn.parameter import Parameter
+
+
+class SincConv(nn.Module):
+    @staticmethod
+    def to_mel(hz):
+        return 2595 * np.log10(1 + hz / 700)
+
+    @staticmethod
+    def to_hz(mel):
+        return 700 * (10 ** (mel / 2595) - 1)
+
+
+    def __init__(self, device,out_channels, kernel_size,in_channels=1,sample_rate=16000,
+                 stride=1, padding=0, dilation=1, bias=False, groups=1):
+
+        super(SincConv,self).__init__()
+
+        if in_channels != 1:
+            
+            msg = "SincConv only support one input channel (here, in_channels = {%i})" % (in_channels)
+            raise ValueError(msg)
+        
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.sample_rate=sample_rate
+
+        # Forcing the filters to be odd (i.e, perfectly symmetrics)
+        if kernel_size%2==0:
+            self.kernel_size=self.kernel_size+1
+
+        self.device=device   
+        self.stride = stride
+        self.padding = padding
+        self.dilation = dilation
+        
+        if bias:
+            raise ValueError('SincConv does not support bias.')
+        if groups > 1:
+            raise ValueError('SincConv does not support groups.')
+        
+        
+        # initialize filterbanks using Mel scale
+        NFFT = 512
+        f=int(self.sample_rate/2)*np.linspace(0,1,int(NFFT/2)+1)
+        fmel=self.to_mel(f)   # Hz to mel conversion
+        fmelmax=np.max(fmel)
+        fmelmin=np.min(fmel)
+        filbandwidthsmel=np.linspace(fmelmin,fmelmax,self.out_channels+1)
+        filbandwidthsf=self.to_hz(filbandwidthsmel)  # Mel to Hz conversion
+        self.mel=filbandwidthsf
+        self.hsupp=torch.arange(-(self.kernel_size-1)/2, (self.kernel_size-1)/2+1)
+        self.band_pass=torch.zeros(self.out_channels,self.kernel_size)
+    
+       
+        
+    def forward(self,x):
+        for i in range(len(self.mel)-1):
+            fmin=self.mel[i]
+            fmax=self.mel[i+1]
+            hHigh=(2*fmax/self.sample_rate)*np.sinc(2*fmax*self.hsupp/self.sample_rate)
+            hLow=(2*fmin/self.sample_rate)*np.sinc(2*fmin*self.hsupp/self.sample_rate)
+            hideal=hHigh-hLow
+            
+            self.band_pass[i,:]=Tensor(np.hamming(self.kernel_size))*Tensor(hideal)
+        
+        band_pass_filter=self.band_pass.to(self.device)
+
+        self.filters = (band_pass_filter).view(self.out_channels, 1, self.kernel_size)
+        
+        return F.conv1d(x, self.filters, stride=self.stride,
+                        padding=self.padding, dilation=self.dilation,
+                         bias=None, groups=1)
+
+
+        
+class Residual_block(nn.Module):
+    def __init__(self, nb_filts, first = False):
+        super(Residual_block, self).__init__()
+        self.first = first
+        
+        if not self.first:
+            self.bn1 = nn.BatchNorm1d(num_features = nb_filts[0])
+        
+        self.lrelu = nn.LeakyReLU(negative_slope=0.3)
+        
+        self.conv1 = nn.Conv1d(in_channels = nb_filts[0],
+			out_channels = nb_filts[1],
+			kernel_size = 3,
+			padding = 1,
+			stride = 1)
+        
+        self.bn2 = nn.BatchNorm1d(num_features = nb_filts[1])
+        self.conv2 = nn.Conv1d(in_channels = nb_filts[1],
+			out_channels = nb_filts[1],
+			padding = 1,
+			kernel_size = 3,
+			stride = 1)
+        
+        if nb_filts[0] != nb_filts[1]:
+            self.downsample = True
+            self.conv_downsample = nn.Conv1d(in_channels = nb_filts[0],
+				out_channels = nb_filts[1],
+				padding = 0,
+				kernel_size = 1,
+				stride = 1)
+            
+        else:
+            self.downsample = False
+        self.mp = nn.MaxPool1d(3)
+        
+    def forward(self, x):
+        identity = x
+        if not self.first:
+            out = self.bn1(x)
+            out = self.lrelu(out)
+        else:
+            out = x
+            
+        out = self.conv1(x)
+        out = self.bn2(out)
+        out = self.lrelu(out)
+        out = self.conv2(out)
+        
+        if self.downsample:
+            identity = self.conv_downsample(identity)
+            
+        out += identity
+        out = self.mp(out)
+        return out
+
+
+
+
+
+class RawNet(nn.Module):
+    def __init__(self, d_args, device):
+        super(RawNet, self).__init__()
+
+        
+        self.device=device
+
+        self.Sinc_conv=SincConv(device=self.device,
+			out_channels = d_args['filts'][0],
+			kernel_size = d_args['first_conv'],
+                        in_channels = d_args['in_channels']
+        )
+        
+        self.first_bn = nn.BatchNorm1d(num_features = d_args['filts'][0])
+        self.selu = nn.SELU(inplace=True)
+        self.block0 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][1], first = True))
+        self.block1 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][1]))
+        self.block2 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][2]))
+        d_args['filts'][2][0] = d_args['filts'][2][1]
+        self.block3 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][2]))
+        self.block4 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][2]))
+        self.block5 = nn.Sequential(Residual_block(nb_filts = d_args['filts'][2]))
+        self.avgpool = nn.AdaptiveAvgPool1d(1)
+
+        self.fc_attention0 = self._make_attention_fc(in_features = d_args['filts'][1][-1],
+            l_out_features = d_args['filts'][1][-1])
+        self.fc_attention1 = self._make_attention_fc(in_features = d_args['filts'][1][-1],
+            l_out_features = d_args['filts'][1][-1])
+        self.fc_attention2 = self._make_attention_fc(in_features = d_args['filts'][2][-1],
+            l_out_features = d_args['filts'][2][-1])
+        self.fc_attention3 = self._make_attention_fc(in_features = d_args['filts'][2][-1],
+            l_out_features = d_args['filts'][2][-1])
+        self.fc_attention4 = self._make_attention_fc(in_features = d_args['filts'][2][-1],
+            l_out_features = d_args['filts'][2][-1])
+        self.fc_attention5 = self._make_attention_fc(in_features = d_args['filts'][2][-1],
+            l_out_features = d_args['filts'][2][-1])
+
+        self.bn_before_gru = nn.BatchNorm1d(num_features = d_args['filts'][2][-1])
+        self.gru = nn.GRU(input_size = d_args['filts'][2][-1],
+			hidden_size = d_args['gru_node'],
+			num_layers = d_args['nb_gru_layer'],
+			batch_first = True)
+
+        
+        self.fc1_gru = nn.Linear(in_features = d_args['gru_node'],
+			out_features = d_args['nb_fc_node'])
+       
+        self.fc2_gru = nn.Linear(in_features = d_args['nb_fc_node'],
+			out_features = d_args['nb_classes'],bias=True)
+			
+       
+        self.sig = nn.Sigmoid()
+        self.logsoftmax = nn.LogSoftmax(dim=1)
+        
+    def forward(self, x, y = None):
+        
+        
+        nb_samp = x.shape[0]
+        len_seq = x.shape[1]
+        x=x.view(nb_samp,1,len_seq)
+        
+        x = self.Sinc_conv(x)    
+        x = F.max_pool1d(torch.abs(x), 3)
+        x = self.first_bn(x)
+        x =  self.selu(x)
+        
+        x0 = self.block0(x)
+        y0 = self.avgpool(x0).view(x0.size(0), -1) # torch.Size([batch, filter])
+        y0 = self.fc_attention0(y0)
+        y0 = self.sig(y0).view(y0.size(0), y0.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x0 * y0 + y0  # (batch, filter, time) x (batch, filter, 1)
+        
+
+        x1 = self.block1(x)
+        y1 = self.avgpool(x1).view(x1.size(0), -1) # torch.Size([batch, filter])
+        y1 = self.fc_attention1(y1)
+        y1 = self.sig(y1).view(y1.size(0), y1.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x1 * y1 + y1 # (batch, filter, time) x (batch, filter, 1)
+
+        x2 = self.block2(x)
+        y2 = self.avgpool(x2).view(x2.size(0), -1) # torch.Size([batch, filter])
+        y2 = self.fc_attention2(y2)
+        y2 = self.sig(y2).view(y2.size(0), y2.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x2 * y2 + y2 # (batch, filter, time) x (batch, filter, 1)
+
+        x3 = self.block3(x)
+        y3 = self.avgpool(x3).view(x3.size(0), -1) # torch.Size([batch, filter])
+        y3 = self.fc_attention3(y3)
+        y3 = self.sig(y3).view(y3.size(0), y3.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x3 * y3 + y3 # (batch, filter, time) x (batch, filter, 1)
+
+        x4 = self.block4(x)
+        y4 = self.avgpool(x4).view(x4.size(0), -1) # torch.Size([batch, filter])
+        y4 = self.fc_attention4(y4)
+        y4 = self.sig(y4).view(y4.size(0), y4.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x4 * y4 + y4 # (batch, filter, time) x (batch, filter, 1)
+
+        x5 = self.block5(x)
+        y5 = self.avgpool(x5).view(x5.size(0), -1) # torch.Size([batch, filter])
+        y5 = self.fc_attention5(y5)
+        y5 = self.sig(y5).view(y5.size(0), y5.size(1), -1)  # torch.Size([batch, filter, 1])
+        x = x5 * y5 + y5 # (batch, filter, time) x (batch, filter, 1)
+
+        x = self.bn_before_gru(x)
+        x = self.selu(x)
+        x = x.permute(0, 2, 1)     #(batch, filt, time) >> (batch, time, filt)
+        self.gru.flatten_parameters()
+        x, _ = self.gru(x)
+        x = x[:,-1,:]
+        x = self.fc1_gru(x)
+        x = self.fc2_gru(x)
+        output=self.logsoftmax(x)
+        print(f"Spec output shape: {output.shape}")
+      
+        return output
+        
+        
+
+    def _make_attention_fc(self, in_features, l_out_features):
+
+        l_fc = []
+        
+        l_fc.append(nn.Linear(in_features = in_features,
+			        out_features = l_out_features))
+
+        
+
+        return nn.Sequential(*l_fc)
+
+
+    def _make_layer(self, nb_blocks, nb_filts, first = False):
+        layers = []
+        #def __init__(self, nb_filts, first = False):
+        for i in range(nb_blocks):
+            first = first if i == 0 else False
+            layers.append(Residual_block(nb_filts = nb_filts,
+				first = first))
+            if i == 0: nb_filts[0] = nb_filts[1]
+            
+        return nn.Sequential(*layers)
+
+    def summary(self, input_size, batch_size=-1, device="cuda", print_fn = None):
+        if print_fn == None: printfn = print
+        model = self
+        
+        def register_hook(module):
+            def hook(module, input, output):
+                class_name = str(module.__class__).split(".")[-1].split("'")[0]
+                module_idx = len(summary)
+                
+                m_key = "%s-%i" % (class_name, module_idx + 1)
+                summary[m_key] = OrderedDict()
+                summary[m_key]["input_shape"] = list(input[0].size())
+                summary[m_key]["input_shape"][0] = batch_size
+                if isinstance(output, (list, tuple)):
+                    summary[m_key]["output_shape"] = [
+						[-1] + list(o.size())[1:] for o in output
+					]
+                else:
+                    summary[m_key]["output_shape"] = list(output.size())
+                    if len(summary[m_key]["output_shape"]) != 0:
+                        summary[m_key]["output_shape"][0] = batch_size
+                        
+                params = 0
+                if hasattr(module, "weight") and hasattr(module.weight, "size"):
+                    params += torch.prod(torch.LongTensor(list(module.weight.size())))
+                    summary[m_key]["trainable"] = module.weight.requires_grad
+                if hasattr(module, "bias") and hasattr(module.bias, "size"):
+                    params += torch.prod(torch.LongTensor(list(module.bias.size())))
+                summary[m_key]["nb_params"] = params
+                
+            if (
+				not isinstance(module, nn.Sequential)
+				and not isinstance(module, nn.ModuleList)
+				and not (module == model)
+			):
+                hooks.append(module.register_forward_hook(hook))
+                
+        device = device.lower()
+        assert device in [
+			"cuda",
+			"cpu",
+		], "Input device is not valid, please specify 'cuda' or 'cpu'"
+        
+        if device == "cuda" and torch.cuda.is_available():
+            dtype = torch.cuda.FloatTensor
+        else:
+            dtype = torch.FloatTensor
+        if isinstance(input_size, tuple):
+            input_size = [input_size]
+        x = [torch.rand(2, *in_size).type(dtype) for in_size in input_size]
+        summary = OrderedDict()
+        hooks = []
+        model.apply(register_hook)
+        model(*x)
+        for h in hooks:
+            h.remove()
+            
+        print_fn("----------------------------------------------------------------")
+        line_new = "{:>20}  {:>25} {:>15}".format("Layer (type)", "Output Shape", "Param #")
+        print_fn(line_new)
+        print_fn("================================================================")
+        total_params = 0
+        total_output = 0
+        trainable_params = 0
+        for layer in summary:
+            # input_shape, output_shape, trainable, nb_params
+            line_new = "{:>20}  {:>25} {:>15}".format(
+				layer,
+				str(summary[layer]["output_shape"]),
+				"{0:,}".format(summary[layer]["nb_params"]),
+			)
+            total_params += summary[layer]["nb_params"]
+            total_output += np.prod(summary[layer]["output_shape"])
+            if "trainable" in summary[layer]:
+                if summary[layer]["trainable"] == True:
+                    trainable_params += summary[layer]["nb_params"]
+            print_fn(line_new)

requirements.txt

@@ -0,0 +1,12 @@
+wget
+timm
+torch
+tensorflow
+moviepy
+librosa
+ffmpeg
+albumentations
+opencv-python
+torchsummary
+onnx
+onnx2pytorch

save_ckpts.py

@@ -0,0 +1,89 @@
+import onnx
+import torch
+import argparse
+import numpy as np
+import torch.nn as nn
+from models.TMC import ETMC
+from models import image
+from onnx2pytorch import ConvertModel
+
+onnx_model = onnx.load('checkpoints\\efficientnet.onnx')
+pytorch_model = ConvertModel(onnx_model)
+
+# Define the audio_args dictionary
+audio_args = {
+    'nb_samp': 64600,
+    'first_conv': 1024,
+    'in_channels': 1,
+    'filts': [20, [20, 20], [20, 128], [128, 128]],
+    'blocks': [2, 4],
+    'nb_fc_node': 1024,
+    'gru_node': 1024,
+    'nb_gru_layer': 3,
+    'nb_classes': 2
+}
+
+
+def get_args(parser):
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--data_dir", type=str, default="datasets/train/fakeavceleb*")
+    parser.add_argument("--LOAD_SIZE", type=int, default=256)
+    parser.add_argument("--FINE_SIZE", type=int, default=224)
+    parser.add_argument("--dropout", type=float, default=0.2)
+    parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
+    parser.add_argument("--hidden", nargs="*", type=int, default=[])
+    parser.add_argument("--hidden_sz", type=int, default=768)
+    parser.add_argument("--img_embed_pool_type", type=str, default="avg", choices=["max", "avg"])
+    parser.add_argument("--img_hidden_sz", type=int, default=1024)
+    parser.add_argument("--include_bn", type=int, default=True)
+    parser.add_argument("--lr", type=float, default=1e-4)
+    parser.add_argument("--lr_factor", type=float, default=0.3)
+    parser.add_argument("--lr_patience", type=int, default=10)
+    parser.add_argument("--max_epochs", type=int, default=500)
+    parser.add_argument("--n_workers", type=int, default=12)
+    parser.add_argument("--name", type=str, default="MMDF")
+    parser.add_argument("--num_image_embeds", type=int, default=1)
+    parser.add_argument("--patience", type=int, default=20)
+    parser.add_argument("--savedir", type=str, default="./savepath/")
+    parser.add_argument("--seed", type=int, default=1)
+    parser.add_argument("--n_classes", type=int, default=2)
+    parser.add_argument("--annealing_epoch", type=int, default=10)
+    parser.add_argument("--device", type=str, default='cpu')
+    parser.add_argument("--pretrained_image_encoder", type=bool, default = False)
+    parser.add_argument("--freeze_image_encoder", type=bool, default = False)
+    parser.add_argument("--pretrained_audio_encoder", type = bool, default=False)
+    parser.add_argument("--freeze_audio_encoder", type = bool, default = False)
+    parser.add_argument("--augment_dataset", type = bool, default = True)
+
+    for key, value in audio_args.items():
+        parser.add_argument(f"--{key}", type=type(value), default=value)
+
+def load_spec_modality_model(args):
+    spec_encoder = image.RawNet(args)
+    ckpt = torch.load('checkpoints\RawNet2.pth', map_location = torch.device('cpu'))
+    spec_encoder.load_state_dict(ckpt, strict = True)
+    spec_encoder.eval()
+    return spec_encoder
+
+
+#Load models.
+parser = argparse.ArgumentParser(description="Train Models")
+get_args(parser)
+args, remaining_args = parser.parse_known_args()
+assert remaining_args == [], remaining_args
+
+spec_model = load_spec_modality_model(args)
+
+print(f"Image model is: {pytorch_model}")
+
+print(f"Audio model is: {spec_model}")
+
+
+PATH = 'checkpoints\\model.pth'
+
+torch.save({
+    'spec_encoder': spec_model.state_dict(),
+    'rgb_encoder': pytorch_model.state_dict()
+}, PATH)
+
+print("Model saved.")

utils/logger.py

@@ -0,0 +1,58 @@
+import logging
+import time
+from datetime import timedelta
+
+
+class LogFormatter:
+    def __init__(self):
+        self.start_time = time.time()
+
+    def format(self, record):
+        elapsed_seconds = round(record.created - self.start_time)
+
+        prefix = "%s - %s - %s" % (
+            record.levelname,
+            time.strftime("%x %X"),
+            timedelta(seconds=elapsed_seconds),
+        )
+        message = record.getMessage()
+        message = message.replace("\n", "\n" + " " * (len(prefix) + 3))
+        return "%s - %s" % (prefix, message)
+
+
+def create_logger(filepath, args):
+    # create log formatter
+    log_formatter = LogFormatter()
+
+    # create file handler and set level to debug
+    file_handler = logging.FileHandler(filepath, "a")
+    file_handler.setLevel(logging.DEBUG)
+    file_handler.setFormatter(log_formatter)
+
+    # create console handler and set level to info
+    console_handler = logging.StreamHandler()
+    console_handler.setLevel(logging.INFO)
+    console_handler.setFormatter(log_formatter)
+
+    # create logger and set level to debug
+    logger = logging.getLogger()
+    logger.handlers = []
+    logger.setLevel(logging.DEBUG)
+    logger.propagate = False
+    logger.addHandler(file_handler)
+    logger.addHandler(console_handler)
+
+    # reset logger elapsed time
+    def reset_time():
+        log_formatter.start_time = time.time()
+
+    logger.reset_time = reset_time
+
+    logger.info(
+        "\n".join(
+            "%s: %s" % (k, str(v))
+            for k, v in sorted(dict(vars(args)).items(), key=lambda x: x[0])
+        )
+    )
+
+    return logger

utils/utils.py

@@ -0,0 +1,55 @@
+import contextlib
+import numpy as np
+import random
+import shutil
+import os
+
+import torch
+
+
+def set_seed(seed):
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed(seed)
+        torch.cuda.manual_seed_all(seed)
+        
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+
+def save_checkpoint(state, is_best, checkpoint_path, filename="checkpoint.pt"):
+    filename = os.path.join(checkpoint_path, filename)
+    torch.save(state, filename)
+    if is_best:
+        shutil.copyfile(filename, os.path.join(checkpoint_path, "model_best.pt"))
+
+
+def load_checkpoint(model, path):
+    best_checkpoint = torch.load(path)
+    model.load_state_dict(best_checkpoint["state_dict"])
+
+def log_metrics(set_name, metrics, logger):
+    logger.info(
+        "{}: Loss: {:.5f} | spec_acc: {:.5f}, rgb_acc: {:.5f}".format(
+            set_name, metrics["loss"], metrics["spec_acc"], metrics["rgb_acc"]
+        )
+    )
+
+
+@contextlib.contextmanager
+def numpy_seed(seed, *addl_seeds):
+    """Context manager which seeds the NumPy PRNG with the specified seed and
+    restores the state afterward"""
+    if seed is None:
+        yield
+        return
+    if len(addl_seeds) > 0:
+        seed = int(hash((seed, *addl_seeds)) % 1e6)
+    state = np.random.get_state()
+    np.random.seed(seed)
+    try:
+        yield
+    finally:
+        np.random.set_state(state)