app.py

import json
from pathlib import Path

import cv2
import numpy as np
import streamlit as st
import torch

from ultralytics import YOLO
from tomato_pipeline import load_classifier, make_transform, classify_crop


# -------------------------
# 🔥 FIX: UPLOAD LIMIT (403 ERROR FIX)
# -------------------------
st.set_option('server.maxUploadSize', 10)  # 10MB


# -------------------------
# CONFIG
# -------------------------
st.set_page_config(
    page_title="Tomato AI Inspector",
    page_icon="🍅",
    layout="wide"
)

st.title("🍅 Tomato AI Quality Inspector")
st.caption("YOLO Detection + EfficientNet Classification")


DETECTOR_PATH = Path("best.pt")
CLASSIFIER_PATH = Path("efficientnet_b0_best.pth")

device = "cuda" if torch.cuda.is_available() else "cpu"


# -------------------------
# LOAD MODELS
# -------------------------
@st.cache_resource
def load_models():
    detector = YOLO(str(DETECTOR_PATH))
    classifier = load_classifier(CLASSIFIER_PATH, device)
    return detector, classifier


detector, classifier = load_models()


# -------------------------
# INPUT
# -------------------------
uploaded = st.file_uploader(
    "Upload Tomato Image",
    type=["jpg", "png", "jpeg"],
    accept_multiple_files=False
)

use_sample = st.button("Use Sample Image")


# -------------------------
# IMAGE LOAD FUNCTION
# -------------------------
def load_image(uploaded_file=None):

    # 👉 SAMPLE IMAGE FALLBACK
    if use_sample:
        try:
            with open("sample.jpg", "rb") as f:
                file_bytes = f.read()
        except:
            st.error("Sample image not found.")
            return None

    # 👉 USER UPLOAD
    elif uploaded_file is not None:
        uploaded_file.file.seek(0)
        file_bytes = uploaded_file.file.read()

        if not file_bytes:
            st.error("Upload failed. Try again.")
            return None

    else:
        return None

    # 👉 DECODE IMAGE
    image_np = cv2.imdecode(
        np.frombuffer(file_bytes, np.uint8),
        cv2.IMREAD_COLOR
    )

    if image_np is None:
        st.error("Invalid image file.")
        return None

    return image_np


# -------------------------
# RUN BUTTON
# -------------------------
run = st.button("Run Detection")


# -------------------------
# INFERENCE
# -------------------------
if run:

    image_np = load_image(uploaded)

    if image_np is None:
        st.warning("Please upload or select an image.")
        st.stop()

    h, w = image_np.shape[:2]
    output = image_np.copy()

    transform = make_transform(224)

    detections = detector.predict(
        source=image_np,
        conf=0.25,
        device=device,
        verbose=False
    )

    results = []
    good_count = 0
    bad_count = 0


    if detections and detections[0].boxes is not None:

        for box in detections[0].boxes:

            x1, y1, x2, y2 = box.xyxy[0].tolist()
            x1, y1, x2, y2 = map(int, [x1, y1, x2, y2])

            crop = image_np[y1:y2, x1:x2]

            if crop.size == 0:
                continue

            label, conf = classify_crop(
                crop,
                classifier,
                transform,
                device,
                ["bad", "good"]
            )

            if label.lower() == "good":
                good_count += 1
                color = (0, 255, 0)
            else:
                bad_count += 1
                color = (0, 0, 255)

            cv2.rectangle(output, (x1, y1), (x2, y2), color, 2)
            cv2.putText(
                output,
                f"{label} {conf:.2f}",
                (x1, y1 - 5),
                cv2.FONT_HERSHEY_SIMPLEX,
                0.6,
                color,
                2
            )

            results.append(label)


    # -------------------------
    # DISPLAY
    # -------------------------
    col1, col2 = st.columns(2)

    with col1:
        st.subheader("Input Image")
        st.image(image_np, channels="BGR", use_container_width=True)

    with col2:
        st.subheader("Detection Result")
        st.image(cv2.cvtColor(output, cv2.COLOR_BGR2RGB), use_container_width=True)

    st.success(f"Total: {len(results)} | Good: {good_count} | Bad: {bad_count}")