app.py

import torch
import gradio as gr
from PIL import Image
from torchvision import transforms
from statistics import mean
Image.MAX_IMAGE_PIXELS = None

def read_img_file(f):
    img = Image.open(f)
    if img.mode != 'RGB':
        img = img.convert('RGB')
    return img

_transform_test_random=transforms.Compose([
        transforms.RandomCrop((256,256)),
        transforms.ToTensor(),
        transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])

_transform_test_random_vit = transforms.Compose([
        transforms.RandomCrop((252,252)),
        transforms.ToTensor(),
        transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])

def detect(img, model_choices):
    if model_choices == "EVA-02 ViT L/14":
        model = torch.load("./model_eva.pth",map_location="cpu").cpu().eval()
        _transform = _transform_test_random_vit
    elif model_choices == "ConvNext Large":
        model = torch.load("./model_convnext.pth",map_location="cpu").cpu().eval()
        _transform = _transform_test_random
    elif model_choices == "EfficientNet-V2 B0":
        model = torch.load("./model_effnet.pth",map_location="cpu").cpu().eval()
        _transform = _transform_test_random

    output = ""
    with torch.inference_mode():
        tmp=[]
        for _ in range(10):
            img_random_crop = _transform(img)
            outputs = model.forward(img_random_crop.unsqueeze(0))
            outputs = torch.sigmoid(outputs).cpu().numpy()
            tmp.append(outputs[0][0])
        output+=f"{str(tmp)}\n"
        output+=f"10 try method: {mean(tmp)}\n"
        # print(tmp)
        # print("10 try method: ", mean(tmp))

    with torch.inference_mode():
        img_crop = _transform(img)
        outputs = model.forward(img_crop.unsqueeze(0))
        outputs = torch.sigmoid(outputs).cpu().numpy()
        output+=f"1 try method: {outputs}\n"
        # print("1 try method: ",outputs)
    return output

model_choices = ["ConvNext Large", "EVA-02 ViT L/14", "EfficientNet-V2 B0"]

descr = f"""
Detecting AutoEncoder is Enough to Catch LDM Generated Images (https://arxiv.org/abs/2411.06441)  
Code at https://github.com/qwertyforce/Detect_LDM_By_Detecting_VAE  
Models at https://huggingface.co/qwertyforce/Detect_LDM_By_Detecting_VAE
"""
demo = gr.Interface(fn=detect, 
        inputs=[gr.Image(type="pil", label="Input Image"),
                 gr.Radio(
                    model_choices,
                    type="value",
                    value="EVA-02 ViT L/14",
                    label="Choose Detector Model",
                )], outputs="text",title = "Detecting AutoEncoder is Enough to Catch LDM Generated Images",description=descr)
demo.launch()