app.py

# -*- coding: utf-8 -*-
"""sign_identifier_gradio_final.ipynb

Gradio interface for image classification using a classmate’s model.
Model: cassieli226/sign-identification-automl
"""

# !pip install autogluon.multimodal gradio huggingface_hub pillow pandas --quiet

import os, pathlib, shutil, zipfile, tempfile, io
import pandas as pd
from PIL import Image
import gradio as gr
import huggingface_hub
from autogluon.multimodal import MultiModalPredictor

# -----------------------------
# Config
# -----------------------------
MODEL_REPO_ID = "cassieli226/sign-identification-automl"
ZIP_FILENAME = "autogluon_predictor_dir.zip"
CACHE_DIR = pathlib.Path("hf_assets")
EXTRACT_DIR = CACHE_DIR / "predictor_native"
MAX_SIZE_MB = 5

# -----------------------------
# Model loading
# -----------------------------
def prepare_predictor_dir() -> str:
    CACHE_DIR.mkdir(parents=True, exist_ok=True)
    local_zip = huggingface_hub.hf_hub_download(
        repo_id=MODEL_REPO_ID,
        filename=ZIP_FILENAME,
        repo_type="model",
        local_dir=str(CACHE_DIR),
        local_dir_use_symlinks=False,
    )
    if EXTRACT_DIR.exists():
        shutil.rmtree(EXTRACT_DIR)
    EXTRACT_DIR.mkdir(parents=True, exist_ok=True)
    with zipfile.ZipFile(local_zip, "r") as zf:
        zf.extractall(str(EXTRACT_DIR))
    contents = list(EXTRACT_DIR.iterdir())
    return str(contents[0]) if (len(contents) == 1 and contents[0].is_dir()) else str(EXTRACT_DIR)

print("Loading predictor...")
PREDICTOR_DIR = prepare_predictor_dir()
PREDICTOR = MultiModalPredictor.load(PREDICTOR_DIR)
print("✅ Model loaded!")

# Try to extract readable class names
try:
    if hasattr(PREDICTOR, "label_generator") and hasattr(PREDICTOR.label_generator, "category_map"):
        CLASS_MAP = {str(k): str(v) for k, v in PREDICTOR.label_generator.category_map.items()}
    else:
        CLASS_MAP = {str(i): str(lbl) for i, lbl in enumerate(PREDICTOR.class_labels)}
except Exception:
    CLASS_MAP = {}
print("Class map:", CLASS_MAP)

# -----------------------------
# Helpers
# -----------------------------
def _pil_to_tmp(img: Image.Image, resize_size=224) -> str:
    img = img.convert("RGB").resize((resize_size, resize_size))
    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
    img.save(tmp.name, format="PNG")
    return tmp.name

def _size_mb_of_png(img: Image.Image) -> float:
    buf = io.BytesIO()
    img.save(buf, format="PNG")
    return buf.tell() / (1024 * 1024)

# -----------------------------
# Inference
# -----------------------------
def predict_image(img, resize_size=224, top_k=3, prob_threshold=0.05):
    if img is None:
        return None, "<div style='color:#b91c1c'>⚠️ Please upload an image.</div>"

    # Validate size
    size_mb = _size_mb_of_png(img)
    if size_mb > MAX_SIZE_MB:
        return None, f"<div style='color:#b91c1c'>⚠️ File too large: {size_mb:.2f} MB (limit {MAX_SIZE_MB} MB).</div>"

    # Preprocess
    img_path = _pil_to_tmp(img, resize_size)
    df = pd.DataFrame({"image": [img_path]})

    # Predict probabilities
    proba_df = PREDICTOR.predict_proba(df)
    probs = proba_df.iloc[0].sort_values(ascending=False)

    # Map numeric indices to actual category names
    probs.index = [CLASS_MAP.get(str(i), str(i)) for i in probs.index]

    # Apply threshold + top-k
    filtered = probs[probs > prob_threshold]
    top = filtered.head(top_k) if not filtered.empty else probs.head(top_k)

    # Top-1
    top_label = top.index[0]
    top_conf = float(top.iloc[0]) * 100

    # HTML result
    html = f"""
    <div style="padding:20px;background:#f0f9ff;border-radius:12px;border-left:5px solid #3b82f6;">
        <h2 style="color:#1e40af;margin:0 0 12px;">🔎 Prediction Results</h2>
        <div style="background:#3b82f6;color:white;padding:15px;border-radius:10px;margin-bottom:15px;text-align:center;">
            <div style="font-size:18px;">Predicted Sign</div>
            <div style="font-size:36px;font-weight:800;letter-spacing:.3px;">{top_label}</div>
            <div style="font-size:16px;opacity:.95;">Confidence: {top_conf:.1f}%</div>
        </div>
        <h4 style="color:#1e40af;margin:10px 0;">Top {len(top)} Predictions</h4>
        <ul style="margin:0 0 10px 18px;color:#111827;">
    """
    for cls, prob in top.items():
        html += f"<li><b>{cls}</b>: {prob*100:.1f}%</li>"
    html += "</ul></div>"

    return img, html

# -----------------------------
# Gradio UI
# -----------------------------
with gr.Blocks(css="""
    .gradio-container { font-family: 'Segoe UI', system-ui, -apple-system, Arial, sans-serif; }
""") as demo:
    gr.HTML(
        "<h1 style='text-align:center;color:#1e40af;'>🚦 Traffic Sign Identifier</h1>"
        "<p style='text-align:center;color:#334155;'>Upload a traffic sign image to see predictions.</p>"
    )

    with gr.Row():
        with gr.Column():
            img_in = gr.Image(type="pil", image_mode="RGB", label="Upload Image", sources=["upload","webcam"])
            resize_size = gr.Slider(64, 512, value=224, step=32, label="Resize Size (px)")
            top_k = gr.Slider(1, 10, value=3, step=1, label="Top-k Predictions")
            prob_threshold = gr.Slider(0.0, 0.9, value=0.05, step=0.01, label="Probability Threshold")
            btn = gr.Button("🔍 Predict", variant="primary")
        with gr.Column():
            orig_out = gr.Image(label="Original Image", image_mode="RGB")
            res_out = gr.HTML(label="Results")

    btn.click(
        fn=predict_image,
        inputs=[img_in, resize_size, top_k, prob_threshold],
        outputs=[orig_out, res_out],
    )

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