app.py

import requests
from PIL import Image
from io import BytesIO
import torch
from torchvision import transforms
from transformers import AutoModelForImageClassification, AutoConfig
import gradio as gr
import spaces
import os

token = os.environ.get("HUGGINGFACE_HUB_TOKEN")

model_id = "thelabel/240903-image-tagging"
config = AutoConfig.from_pretrained(model_id, token=token)
model = AutoModelForImageClassification.from_pretrained(model_id, token=token)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

image_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]),
])

def load_image_from_url(url):
    try:
        response = requests.get(url.strip(), timeout=10)
        response.raise_for_status()
        return Image.open(BytesIO(response.content)).convert("RGB")
    except Exception:
        return None

@spaces.GPU
def predict_tags(image_url, threshold=0.5):
    image = load_image_from_url(image_url)
    if image is None:
        return None, "Could not load image."

    image_tensor = image_transform(image).unsqueeze(0).to(device)
    with torch.no_grad():
        logits = model(image_tensor).logits
        probs = torch.sigmoid(logits).squeeze()

    results = [
        (config.idx_to_label[str(i)], float(probs[i]))
        for i in range(len(probs))
        if probs[i] >= threshold
    ]
    results.sort(key=lambda x: x[1], reverse=True)
    return results, None

def gradio_predict(urls, threshold):
    url_list = [u.strip() for u in urls.split(",") if u.strip()]
    output = []

    for url in url_list:
        tags, error = predict_tags(url, threshold)
        if error or not tags:
            output.append({
                "image_url": url,
                "error": error or "No tags above threshold."
            })
        else:
            top_tag, top_score = tags[0]
            output.append({
                "image_url": url,
                "tag_name": top_tag,
                "tag_score": round(top_score, 4)
            })

    return str(output)  # Return as string for textbox display

demo = gr.Interface(
    fn=gradio_predict,
    inputs=[
        gr.Textbox(label="Image URL(s) (comma-separated)"),
        gr.Slider(0, 1, value=0.5, step=0.01, label="Threshold"),
    ],
    outputs=gr.Textbox(label="Tags"),
    title="Batch Image Tagging with ViT",
    description="Paste one or more image URLs separated by commas to get predicted tags using thelabel/240903-image-tagging model.",
)

if __name__ == "__main__":
    demo.launch()