app.py

import torch
import streamlit as st
import requests
from pathlib import Path

# Define model path and URL
MODEL_PATH = "pytorch_model.pth"
MODEL_URL = "https://huggingface.co/zongzhuofan/co-detr-vit-large-coco/resolve/main/pytorch_model.pth"

# Download model if not exists
@st.cache_resource
def download_model():
    if not Path(MODEL_PATH).exists():
        with st.spinner("Downloading model... This might take a few minutes..."):
            response = requests.get(MODEL_URL, stream=True)
            with open(MODEL_PATH, "wb") as f:
                for chunk in response.iter_content(chunk_size=8192):
                    if chunk:
                        f.write(chunk)
    return MODEL_PATH

# Load the model
model = YourModelClass()
model_path = download_model()
model.load_state_dict(torch.load(model_path, map_location='cpu'))
model.eval()

st.title("Co-DETR Model")

uploaded_file = st.file_uploader("Upload an image", type=["jpg", "png"])
if uploaded_file is not None:
    # Preprocess the image
    input_data = preprocess_image(uploaded_file)
    with torch.no_grad():
        output = model(input_data)
    # Postprocess output if necessary
    st.write("Output:", output)