Spaces:
Sleeping
Sleeping
File size: 2,965 Bytes
82eb50a a00c2f7 82eb50a a00c2f7 04247ac 6c35a64 a00c2f7 |
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 |
import gradio as gr
from transformers import pipeline, ViTForImageClassification, ViTImageProcessor
import numpy as np
from PIL import Image
import warnings
import logging
from pytorch_grad_cam import run_dff_on_image, GradCAM
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
from pytorch_grad_cam.utils.image import show_cam_on_image
import torch
from face_grab import FaceGrabber
from gradcam import GradCam
from torchvision import transforms
logging.basicConfig(level=logging.INFO)
model = ViTForImageClassification.from_pretrained("ongkn/emikes-classifier")
processor = ViTImageProcessor.from_pretrained("ongkn/emikes-classifier")
faceGrabber = FaceGrabber()
gradCam = GradCam()
targetsForGradCam = [ClassifierOutputTarget(gradCam.category_name_to_index(model, "emi")),
ClassifierOutputTarget(gradCam.category_name_to_index(model, "kes"))]
targetLayerDff = model.vit.layernorm
targetLayerGradCam = model.vit.encoder.layer[-2].output
def classify_image(input):
face = faceGrabber.grab_faces(np.array(input))
if face is None:
return "No face detected", 0, input
face = Image.fromarray(face)
faceResized = face.resize((224, 224))
tensorResized = transforms.ToTensor()(faceResized)
dffImage = run_dff_on_image(model=model,
target_layer=targetLayerDff,
classifier=model.classifier,
img_pil=faceResized,
img_tensor=tensorResized,
reshape_transform=gradCam.reshape_transform_vit_huggingface,
n_components=5,
top_k=10
)
result = gradCam.get_top_category(model, tensorResized)
cls = result[0]["label"]
clsIdx = gradCam.category_name_to_index(model, cls)
clsTarget = ClassifierOutputTarget(clsIdx)
gradCamImage = gradCam.run_grad_cam_on_image(model=model,
target_layer=targetLayerGradCam,
targets_for_gradcam=[clsTarget],
input_tensor=tensorResized,
input_image=faceResized,
reshape_transform=gradCam.reshape_transform_vit_huggingface)
return result[0]["label"], result[0]["score"], face, dffImage, gradCamImage
iface = gr.Interface(
fn=classify_image,
inputs="image",
outputs=["text", "number", "image", "image", "image"],
title="Emikes Classifier",
description=f"Takes in a (224, 224) image and outputs a class: {'emi', 'kes'}, along with a GradCam/DFF explanation. Face detection, cropping, and resizing are done internally. Uploaded images are not stored by us, but may be stored by HF. Refer to their [privacy policy](https://huggingface.co/privacy) for details."
)
iface.launch() |