Spaces:
Running
Running
File size: 8,183 Bytes
06d8add 023b809 17eb7fe 06d8add 1aa551a 06d8add 3e69f0d 7c5fb79 54755ff 06d8add 59b5c95 06d8add 2464264 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 |
import os
import sys
os.system('git clone https://github.com/facebookresearch/av_hubert.git')
os.chdir('/home/user/app/av_hubert')
os.system('git submodule init')
os.system('git submodule update')
os.chdir('/home/user/app/av_hubert/fairseq')
os.system('pip install ./')
os.system('pip install scipy')
os.system('pip install sentencepiece')
os.system('pip install python_speech_features')
os.system('pip install scikit-video')
os.system('pip install transformers')
os.system('pip install gradio==3.12')
os.system('pip install numpy==1.23.3')
os.chdir('/home/user/app')
os.makedirs("./result", exist_ok = True)
os.makedirs("./video/và/test", exist_ok = True)
# sys.path.append('/home/user/app/av_hubert')
sys.path.append('/home/user/app/av_hubert/avhubert')
print(sys.path)
print(os.listdir())
print(sys.argv, type(sys.argv))
sys.argv.append('dummy')
import dlib, cv2, os
import numpy as np
import skvideo
import skvideo.io
from tqdm import tqdm
from preparation.align_mouth import landmarks_interpolate, crop_patch, write_video_ffmpeg
from base64 import b64encode
import torch
import cv2
import tempfile
from argparse import Namespace
import fairseq
from fairseq import checkpoint_utils, options, tasks, utils
from fairseq.dataclass.configs import GenerationConfig
from huggingface_hub import hf_hub_download
import gradio as gr
from pytube import YouTube
# os.chdir('/home/user/app/av_hubert/avhubert')
user_dir = "/home/user/app/av_hubert/avhubert"
utils.import_user_module(Namespace(user_dir=user_dir))
data_dir = "/home/user/app/video"
# ckpt_path = hf_hub_download('vumichien/AV-HuBERT', 'model.pt')
face_detector_path = "/home/user/app/mmod_human_face_detector.dat"
face_predictor_path = "/home/user/app/shape_predictor_68_face_landmarks.dat"
mean_face_path = "/home/user/app/20words_mean_face.npy"
mouth_roi_path = "/home/user/app/roi.mp4"
output_video_path = "/home/user/app/video/và/test"
modalities = ["video"]
gen_subset = "test"
gen_cfg = GenerationConfig(beam=20)
# models, saved_cfg, task = checkpoint_utils.load_model_ensemble_and_task([ckpt_path])
# models = [model.eval().cuda() if torch.cuda.is_available() else model.eval() for model in models]
# saved_cfg.task.modalities = modalities
# saved_cfg.task.data = data_dir
# saved_cfg.task.label_dir = data_dir
# task = tasks.setup_task(saved_cfg.task)
# generator = task.build_generator(models, gen_cfg)
def get_youtube(video_url):
yt = YouTube(video_url)
abs_video_path = yt.streams.filter(progressive=True, file_extension='mp4').order_by('resolution').desc().first().download()
print("Success download video")
print(abs_video_path)
return abs_video_path
import dlib, cv2, os
import numpy as np
import skvideo
import skvideo.io
from tqdm import tqdm
from preparation.align_mouth import landmarks_interpolate, crop_patch, write_video_ffmpeg
from IPython.display import HTML
from base64 import b64encode
import numpy as np
def convert_bgr2gray(data):
# np.stack(배열_1, 배열_2, axis=0): 지정한 axis를 완전히 새로운 axis로 생각
return np.stack([cv2.cvtColor(_, cv2.COLOR_BGR2GRAY) for _ in data], axis=0)
def save2npz(filename, data=None):
"""save2npz.
:param filename: str, the fileanme where the data will be saved.
:param data: ndarray, arrays to save to the file.
"""
assert data is not None, "data is {}".format(data)
if not os.path.exists(os.path.dirname(filename)):
os.makedirs(os.path.dirname(filename))
np.savez_compressed(filename, data=data)
def detect_landmark(image, detector, predictor):
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
face_locations = detector(gray, 1)
coords = None
for (_, face_location) in enumerate(face_locations):
if torch.cuda.is_available():
rect = face_location.rect
else:
rect = face_location
shape = predictor(gray, rect)
coords = np.zeros((68, 2), dtype=np.int32)
for i in range(0, 68):
coords[i] = (shape.part(i).x, shape.part(i).y)
return coords
def preprocess_video(input_video_path):
if torch.cuda.is_available():
detector = dlib.cnn_face_detection_model_v1(face_detector_path)
else:
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(face_predictor_path)
STD_SIZE = (256, 256)
mean_face_landmarks = np.load(mean_face_path)
stablePntsIDs = [33, 36, 39, 42, 45]
videogen = skvideo.io.vread(input_video_path)
frames = np.array([frame for frame in videogen])
landmarks = []
for frame in tqdm(frames):
landmark = detect_landmark(frame, detector, predictor)
landmarks.append(landmark)
preprocessed_landmarks = landmarks_interpolate(landmarks)
rois = crop_patch(input_video_path, preprocessed_landmarks, mean_face_landmarks, stablePntsIDs, STD_SIZE,
window_margin=12, start_idx=48, stop_idx=68, crop_height=96, crop_width=96)
rois_gray=convert_bgr2gray(rois)
save2npz(output_video_path, data=rois_gray)
write_video_ffmpeg(rois, mouth_roi_path, "/usr/bin/ffmpeg")
return mouth_roi_path
def predict(process_video):
os.chdir('/home/user/app')
try:
os.system('bash ./TestVisual.sh')
pred=[]
fi = open("/home/user/app/result/ho.txt", 'r')
pred = fi.readlines()
except IndexError:
pred=['error']
return pred[-1]
# ---- Gradio Layout -----
youtube_url_in = gr.Textbox(label="Youtube url", lines=1, interactive=True)
video_in = gr.Video(label="Video đầu vào", mirror_webcam=False, interactive=True)
video_out = gr.Video(label="Video vùng môi", mirror_webcam=False, interactive=True)
demo = gr.Blocks()
demo.encrypt = False
text_output = gr.Textbox()
with demo:
# gr.Markdown('''
# <div>
# <h1 style='text-align: center'>Speech Recognition from Visual Lip Movement by Audio-Visual Hidden Unit BERT Model (AV-HuBERT)</h1>
# This space uses AV-HuBERT models from <a href='https://github.com/facebookresearch' target='_blank'><b>Meta Research</b></a> to recoginze the speech from Lip Movement 🤗
# <figure>
# <img src="https://huggingface.co/vumichien/AV-HuBERT/resolve/main/lipreading.gif" alt="Audio-Visual Speech Recognition">
# <figcaption> Speech Recognition from visual lip movement
# </figcaption>
# </figure>
# </div>
# ''')
# with gr.Row():
# gr.Markdown('''
# ### Reading Lip movement with youtube link using Avhubert
# ##### Step 1a. Download video from youtube (Note: the length of video should be less than 10 seconds if not it will be cut and the face should be stable for better result)
# ##### Step 1b. You also can upload video directly
# ##### Step 2. Generating landmarks surrounding mouth area
# ##### Step 3. Reading lip movement.
# ''')
with gr.Row():
gr.Markdown('''
### You can test by following examples:
''')
examples = gr.Examples(examples=
[ "https://www.youtube.com/watch?v=ZXVDnuepW2s",
"https://www.youtube.com/watch?v=X8_glJn1B8o",
"https://www.youtube.com/watch?v=80yqL2KzBVw"],
label="Examples", inputs=[youtube_url_in])
with gr.Column():
youtube_url_in.render()
download_youtube_btn = gr.Button("Download Youtube video")
download_youtube_btn.click(get_youtube, [youtube_url_in], [
video_in])
print(video_in)
with gr.Row():
video_in.render()
video_out.render()
with gr.Row():
detect_landmark_btn = gr.Button("Phát hiện mốc/cắt môi")
detect_landmark_btn.click(preprocess_video, [video_in], [
video_out])
predict_btn = gr.Button("Dự đoán")
predict_btn.click(predict, [video_out], [
text_output])
with gr.Row():
# video_lip = gr.Video(label="Audio Visual Video", mirror_webcam=False)
text_output.render()
demo.launch(debug=True) |