app.py

import gradio as gr
from transformers import CLIPProcessor, CLIPModel
import torch
import requests
from PIL import Image
from io import BytesIO

fashion_items = ['top', 'trousers', 'bottom', 'jumper']

# Load model and processor - using standard CLIP model instead
model_name = "openai/clip-vit-base-patch32"
model = CLIPModel.from_pretrained(model_name)
processor = CLIPProcessor.from_pretrained(model_name)

# CLIP processes text and images together, so no need for separate text preprocessing

# Prediction function
def predict_from_url(url):
    # Check if the URL is empty
    if not url:
        return {"Error": "Please input a URL"}
    
    try:
        image = Image.open(BytesIO(requests.get(url).content))
    except Exception as e:
        return {"Error": f"Failed to load image: {str(e)}"}
    
    inputs = processor(images=image, text=fashion_items, return_tensors="pt", padding=True)
    
    with torch.no_grad():
        outputs = model(**inputs)
        logits_per_image = outputs.logits_per_image
        text_probs = logits_per_image.softmax(dim=-1)
        
    return {fashion_items[i]: float(text_probs[0, i]) for i in range(len(fashion_items))}

# Gradio interface
demo = gr.Interface(
    fn=predict_from_url, 
    inputs=gr.Textbox(label="Enter Image URL"),
    outputs=gr.Label(label="Classification Results"), 
    title="Fashion Item Classifier",
    allow_flagging="never"
)

# Launch the interface
demo.launch()