import gradio as gr
import numpy as np
from PIL import Image
import tensorflow as tf
from typing import List, Dict, Any
import io

# Labels must mirror src/classification-model/index.ts
LABELS: List[str] = [
    "battery",
    "biological",
    "brown-glass",
    "cardboard",
    "clothes",
    "green-glass",
    "metal",
    "paper",
    "plastic",
    "shoes",
    "trash",
    "white-glass",
]


def _load_image_to_rgb(image: Image.Image) -> np.ndarray:
    if image.mode != "RGB":
        image = image.convert("RGB")
    return np.asarray(image)


def _resize_224(img_rgb: np.ndarray) -> np.ndarray:
    im = Image.fromarray(img_rgb)
    im = im.resize((224, 224), Image.NEAREST)
    return np.asarray(im)


def _preprocess(image: Image.Image) -> np.ndarray:
    rgb = _load_image_to_rgb(image)
    rgb224 = _resize_224(rgb)
    # shape [1,224,224,3], float32 in 0..255
    arr = rgb224.astype("float32")
    return np.expand_dims(arr, axis=0)


class PreTrainedModel:
    def __init__(self, model_path: str = "model/model_resnet50.keras") -> None:
        self.model = tf.keras.models.load_model(model_path)

    def predict_image(self, image: Image.Image) -> Dict[str, float]:
        x = _preprocess(image)
        preds = self.model.predict(x)
        if isinstance(preds, (list, tuple)):
            preds = preds[0]
        probs = np.asarray(preds).squeeze().tolist()

        return {label: score for label, score in zip(LABELS, probs)}


model = PreTrainedModel()


def predict(image):
    predictions = model.predict_image(image)

    probs_percent = {label: round(p * 100, 2)
                     for label, p in predictions.items()}

    max_label = max(probs_percent, key=probs_percent.get)

    return {
        "label": max_label,
        "percentage": probs_percent[max_label],
        "probabilities": probs_percent,
    }


iface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.JSON(),
    title="Waste Classification",
    description="Upload an image of waste to classify it.",
)

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