not-lain/deepfake-detection

1import gradio as gr2import torch3import torch.nn.functional as F4from facenet_pytorch import MTCNN, InceptionResnetV15import os6import numpy as np7from PIL import Image8import zipfile9import cv210from pytorch_grad_cam import GradCAM11from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget12from pytorch_grad_cam.utils.image import show_cam_on_image13from transformers import pipeline14with zipfile.ZipFile("examples.zip","r") as zip_ref:15    zip_ref.extractall(".")16 17pipe = pipeline(model="not-lain/deepfake",trust_remote_code=True)18 19EXAMPLES_FOLDER = 'examples'20examples_names = os.listdir(EXAMPLES_FOLDER)21examples = []22for example_name in examples_names:23    example_path = os.path.join(EXAMPLES_FOLDER, example_name)24    label = example_name.split('_')[0]25    example = {26        'path': example_path,27        'label': label28    }29    examples.append(example)30np.random.shuffle(examples) # shuffle31 32def predict(input_image:Image.Image, true_label:str):33    out = pipe.predict(input_image)34    confidences,face_with_mask = out["confidences"], out["face_with_mask"]35    return confidences, true_label, face_with_mask36 37interface = gr.Interface(38    fn=predict,39    inputs=[40        gr.Image(label="Input Image", type="filepath"),41        "text"42    ],43    outputs=[44        gr.Label(label="Class"),45        "text",46        gr.Image(label="Face with Explainability")47    ],48    examples=[[examples[i]["path"], examples[i]["label"]] for i in range(10)]49).launch()