"""Gradio Space — Paint Defect Detector."""
from __future__ import annotations

import os
import sys
import tempfile
from pathlib import Path

import cv2
import gradio as gr
import numpy as np
import torch

# ── paths ──────────────────────────────────────────────────────────────────
ROOT = Path(__file__).resolve().parent
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))

# inline constants so the app works even if src imports have issues
DEFECT_THRESHOLD = float(os.getenv("PDD_DEFECT_THRESHOLD", 0.55))
PANEL_DEFECT_RATIO = 0.005
IMG_SIZE = 384
PATCH_SIZE = 512
PATCH_STRIDE = 256
BACKBONE = os.getenv("PDD_BACKBONE", "tf_efficientnetv2_s.in21k_ft_in1k")

from src.infer import load_model, predict_image, render_visualization

# ── model (lazy) ───────────────────────────────────────────────────────────
_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
_model = None

CHECKPOINT = ROOT / "checkpoints" / "best.pt"


def _get_model():
    global _model
    if _model is None:
        if not CHECKPOINT.exists():
            raise gr.Error(
                "No trained checkpoint found at checkpoints/best.pt. "
                "Please train the model first and upload the checkpoint."
            )
        _model = load_model(CHECKPOINT, device=_device)
    return _model


# ── inference ──────────────────────────────────────────────────────────────

def run_inference(image: np.ndarray, vin: str, threshold: float) -> tuple:
    if image is None:
        return None, "⚠️ Please upload an image.", ""

    # Convert RGB (Gradio) → BGR (OpenCV)
    bgr = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

    try:
        model = _get_model()
    except gr.Error as e:
        return None, str(e), ""

    result = predict_image(bgr, model, _device, threshold=threshold)
    vis_bgr = render_visualization(result)
    vis_rgb = cv2.cvtColor(vis_bgr, cv2.COLOR_BGR2RGB)

    # ── verdict text ──────────────────────────────────────────────────────
    verdict = "🔴  DEFECT DETECTED" if result["is_defect"] else "🟢  NO DEFECT — OK"
    vin_line = f"**VIN:** {vin.strip()}\n\n" if vin.strip() else ""
    summary = (
        f"{vin_line}"
        f"**Verdict:** {verdict}\n\n"
        f"**Defect ratio:** {result['defect_ratio']*100:.2f}%\n\n"
        f"**Max patch probability:** {result['max_prob']:.3f}\n\n"
        f"**Defect regions found:** {len(result['boxes'])}\n\n"
        f"**Panel size:** {result['panel_size']['w']} × {result['panel_size']['h']} px"
    )

    # ── boxes table ───────────────────────────────────────────────────────
    if result["boxes"]:
        rows = "\n".join(
            f"| {i+1} | {b['x']},{b['y']} | {b['w']}×{b['h']} | {b['confidence']:.3f} |"
            for i, b in enumerate(result["boxes"])
        )
        table = (
            "### Defect Regions\n"
            "| # | Position (x,y) | Size (w×h) | Confidence |\n"
            "|---|----------------|------------|------------|\n"
            + rows
        )
    else:
        table = ""

    return vis_rgb, summary, table


# ── UI ─────────────────────────────────────────────────────────────────────

DESCRIPTION = """
# 🚗 Paint Defect Detector

Upload a photo of a car body panel to detect paint defects using a sliding-window
**EfficientNetV2-S** classifier. The model returns a heatmap overlay with bounding
boxes around defective regions.

> **Note:** A trained checkpoint (`checkpoints/best.pt`) must be present.
"""

with gr.Blocks(title="Paint Defect Detector", theme=gr.themes.Soft()) as demo:
    gr.Markdown(DESCRIPTION)

    with gr.Row():
        with gr.Column(scale=1):
            img_input = gr.Image(label="Car Body Panel Photo", type="numpy")
            vin_input = gr.Textbox(label="VIN (optional)", placeholder="e.g. XTA210930Y2837465")
            threshold = gr.Slider(
                minimum=0.1, maximum=0.9, value=DEFECT_THRESHOLD, step=0.05,
                label="Defect Threshold",
                info="Patch probability above this value is marked as defective."
            )
            run_btn = gr.Button("🔍 Analyze", variant="primary")

        with gr.Column(scale=1):
            img_output = gr.Image(label="Visualization (Heatmap + Bounding Boxes)", type="numpy")
            verdict_md = gr.Markdown(label="Result")
            table_md = gr.Markdown(label="Defect Regions")

    run_btn.click(
        fn=run_inference,
        inputs=[img_input, vin_input, threshold],
        outputs=[img_output, verdict_md, table_md],
    )

    gr.Examples(
        examples=[],
        inputs=[img_input],
        label="Examples"
    )

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