import streamlit as st
import torch
import open_clip
from PIL import Image
from classifier import few_shot_fault_classification

# Load lightweight CLIP model
device = "cuda" if torch.cuda.is_available() else "cpu"
model, _, preprocess = open_clip.create_model_and_transforms('RN50', pretrained='openai')
model = model.to(device)
model.eval()

st.title("🛠️ Few-Shot Fault Detection (Industrial Quality Control)")

st.markdown("Upload **10 Nominal Images**, **10 Defective Images**, and one or more **Test Images** to classify.")

col1, col2 = st.columns(2)
with col1:
    nominal_files = st.file_uploader("Upload Nominal Images", type=["png", "jpg", "jpeg"], accept_multiple_files=True)
with col2:
    defective_files = st.file_uploader("Upload Defective Images", type=["png", "jpg", "jpeg"], accept_multiple_files=True)

test_files = st.file_uploader("Upload Test Images", type=["png", "jpg", "jpeg"], accept_multiple_files=True)

if st.button("Classify Test Images"):
    if len(nominal_files) < 1 or len(defective_files) < 1 or len(test_files) < 1:
        st.warning("Please upload at least 1 image in each category.")
    else:
        st.info("Running classification...")

        nominal_imgs = [preprocess(Image.open(f).convert("RGB")).unsqueeze(0) for f in nominal_files]
        defective_imgs = [preprocess(Image.open(f).convert("RGB")).unsqueeze(0) for f in defective_files]
        test_imgs = [preprocess(Image.open(f).convert("RGB")).unsqueeze(0) for f in test_files]

        results = few_shot_fault_classification(
            model=model,
            test_images=[img.squeeze(0) for img in test_imgs],
            test_image_filenames=[f.name for f in test_files],
            nominal_images=[img.squeeze(0) for img in nominal_imgs],
            nominal_descriptions=[f.name for f in nominal_files],
            defective_images=[img.squeeze(0) for img in defective_imgs],
            defective_descriptions=[f.name for f in defective_files],
            num_few_shot_nominal_imgs=len(nominal_files),
            device=device
        )

        for res in results:
            st.write(f"**{res['image_path']}** ➜ {res['classification_result']} "
                     f"(Nominal: {res['non_defect_prob']}, Defective: {res['defect_prob']})")