app.py

import cv2
import gradio as gr
import os
import requests
from PIL import Image
import numpy as np
import torch
from torch.autograd import Variable
from torchvision import transforms
import torch.nn.functional as F

# Automatically download required files
# 1. data_loader_cache.py from GitHub
if not os.path.exists("data_loader_cache.py"):
    print("Downloading data_loader_cache.py...")
    try:
        response = requests.get("https://raw.githubusercontent.com/xuebinqin/DIS/main/DIS/IS-Net/data_loader_cache.py")
        response.raise_for_status()
        with open("data_loader_cache.py", "wb") as f:
            f.write(response.content)
    except requests.RequestException as e:
        print(f"Failed to download data_loader_cache.py: {e}")
        raise

# 2. models.py from GitHub
if not os.path.exists("models.py"):
    print("Downloading models.py...")
    try:
        response = requests.get("https://raw.githubusercontent.com/xuebinqin/DIS/main/DIS/IS-Net/models.py")
        response.raise_for_status()
        with open("models.py", "wb") as f:
            f.write(response.content)
    except requests.RequestException as e:
        print(f"Failed to download models.py: {e}")
        raise

# 3. isnet.pth from Hugging Face Git LFS (direct URL from screenshot)
if not os.path.exists("saved_models"):
    os.makedirs("saved_models")
isnet_path = "saved_models/isnet.pth"
if not os.path.exists(isnet_path):
    print("Downloading isnet.pth from Hugging Face Git LFS...")
    try:
        lfs_url = "https://cdn-lfs.huggingface.co/repos/e0/a8/e0a889743a78391b48db7c4c0b4de1963ee320cb10934c75a32481dc5af9c61/e0a889743a78391b48db7c4c0b4de1963ee320cb10934c75a32481dc5af9c61?download=true"
        response = requests.get(lfs_url, stream=True)
        response.raise_for_status()
        with open(isnet_path, "wb") as f:
            for chunk in response.iter_content(chunk_size=8192):
                if chunk:
                    f.write(chunk)
    except requests.RequestException as e:
        print(f"Failed to download isnet.pth: {e}")
        raise

# Project imports
from data_loader_cache import normalize, im_reader, im_preprocess
from models import *

# Helpers
device = 'cpu'

class GOSNormalize(object):
    def __init__(self, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]):
        self.mean = mean
        self.std = std

    def __call__(self, image):
        image = normalize(image, self.mean, self.std)
        return image

transform = transforms.Compose([GOSNormalize([0.5, 0.5, 0.5], [1.0, 1.0, 1.0])])

def load_image(im_path, hypar):
    im = im_reader(im_path)
    im, im_shp = im_preprocess(im, hypar["cache_size"])
    im = torch.divide(im, 255.0)
    shape = torch.from_numpy(np.array(im_shp))
    return transform(im).unsqueeze(0), shape.unsqueeze(0)

def build_model(hypar, device):
    net = hypar["model"]
    net.to(device)
    if hypar["restore_model"]:
        net.load_state_dict(torch.load(os.path.join(hypar["model_path"], hypar["restore_model"]), map_location=device))
    net.eval()
    return net

def predict(net, inputs_val, shapes_val, hypar, device):
    net.eval()
    inputs_val = inputs_val.type(torch.FloatTensor).to(device)
    with torch.no_grad():
        inputs_val_v = Variable(inputs_val)
        ds_val = net(inputs_val_v)[0]
        pred_val = ds_val[0][0, :, :, :]
        pred_val = torch.squeeze(F.upsample(torch.unsqueeze(pred_val, 0), (shapes_val[0][0], shapes_val[0][1]), mode='bilinear'))
        ma = torch.max(pred_val)
        mi = torch.min(pred_val)
        pred_val = (pred_val - mi) / (ma - mi)
        return (pred_val.cpu().numpy() * 255).astype(np.uint8)

# Set Parameters
hypar = {
    "model_path": "saved_models",
    "restore_model": "isnet.pth",
    "cache_size": [512, 512],
    "input_size": [512, 512],
    "crop_size": [512, 512],
    "model": ISNetDIS()
}

# Build Model
net = build_model(hypar, device)

def inference(image):
    image_path = image
    image_tensor, orig_size = load_image(image_path, hypar)
    mask = predict(net, image_tensor, orig_size, hypar, device)
    pil_mask = Image.fromarray(mask).convert('L')
    im_rgb = Image.open(image).convert("RGB")
    im_rgba = im_rgb.copy()
    im_rgba.putalpha(pil_mask)
    return [im_rgba, pil_mask]

title = "Dichotomous Image Segmentation"
description = "Upload an image to remove its background."

interface = gr.Interface(
    fn=inference,
    inputs=gr.Image(type='filepath'),
    outputs=[gr.Image(type='filepath', format="png"), gr.Image(type='filepath', format="png")],
    title=title,
    description=description,
    flagging_mode="never",
    cache_mode="lazy"
).launch()