app.py

import os, zipfile, shutil
import torch, clip
from PIL import Image
import gradio as gr
import numpy as np

# Load CLIP model (CPU mode works on Spaces)
device = "cpu"
model, preprocess = clip.load("ViT-B/32", device=device)

# Your 8 landmark prompts
landmarks = [
    "Starfield library",
    "Statue of King Sejong",
    "Cheomseongdae",
    "N Seoul Tower",
    "63 Building",
    "Jongno Tower",
    "Gocheok Sky Dome",
    "Myeongdong Cathedral"
]
text_tokens = clip.tokenize(landmarks).to(device)


def classify_images(zip_file, threshold=0.3):
    # Prepare temp folders
    work_dir = "workspace"
    images_dir = os.path.join(work_dir, "images")
    matched_dir = os.path.join(work_dir, "matched")

    shutil.rmtree(work_dir, ignore_errors=True)
    os.makedirs(images_dir, exist_ok=True)
    os.makedirs(matched_dir, exist_ok=True)

    # Extract uploaded zip
    with zipfile.ZipFile(zip_file, 'r') as zip_ref:
        zip_ref.extractall(images_dir)

    results = []
    # Loop over images
    for f in os.listdir(images_dir):
        if f.lower().endswith(('.jpg', '.jpeg', '.png')):
            path = os.path.join(images_dir, f)
            try:
                image = preprocess(Image.open(path)).unsqueeze(0).to(device)
                with torch.no_grad():
                    logits_per_image, _ = model(image, text_tokens)
                    probs = logits_per_image.softmax(dim=-1).cpu().numpy()[0]

                best_idx = int(np.argmax(probs))
                best_label = landmarks[best_idx]
                best_conf = float(probs[best_idx])

                if best_conf > threshold:
                    shutil.copy(path, os.path.join(matched_dir, f))
                    results.append([f, best_label, round(best_conf, 2)])
            except Exception as e:
                results.append([f, f"Error: {e}", 0.0])

    # Create zip of matched images
    matched_zip = os.path.join(work_dir, "matched_images.zip")
    shutil.make_archive(matched_zip[:-4], 'zip', matched_dir)

    return results, matched_zip


demo = gr.Interface(
    fn=classify_images,
    inputs=[
        gr.File(label="Upload .zip of images", file_types=[".zip"]),
        gr.Slider(0.0, 1.0, value=0.3, label="Confidence Threshold")
    ],
    outputs=[
        gr.Dataframe(headers=["Filename", "Predicted Landmark", "Confidence"], label="Matching Images"),
        gr.File(label="Download Matched Images (.zip)")
    ],
    title="Landmark Classifier (CLIP)",
    description="Upload a .zip folder of images. The app finds images of these 8 landmarks:\n"
                + ", ".join(landmarks)
)

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