model.py

import os
import json
import torch
import numpy as np
import tensorflow as tf
from PIL import Image
from torchvision import models, transforms

# ================= PATHS =================

BASE_DIR = os.path.dirname(__file__)
MODELS_DIR = os.path.join(BASE_DIR, "models")
LABELS_DIR = os.path.join(BASE_DIR, "labels")

# ================= TF CUSTOM OBJECTS =================

@tf.keras.utils.register_keras_serializable()
class FixedDropout(tf.keras.layers.Dropout):
    def __init__(self, rate, noise_shape=None, seed=None, **kwargs):
        super().__init__(rate, noise_shape=noise_shape, seed=seed, **kwargs)

# Dummy class to satisfy EfficientNet deserialization
@tf.keras.utils.register_keras_serializable()
class EfficientNetB3(tf.keras.Model):
    pass

# ================= INPUT SIZE PER MODEL =================

KERAS_INPUT_SIZES = {
    "corn": 300,
}

# ================= IMAGE PREPROCESS =================

def preprocess_pytorch(img, size=224):
    transform = transforms.Compose([
        transforms.Resize((size, size)),
        transforms.ToTensor(),
        transforms.Normalize(
            mean=[0.485, 0.456, 0.406],
            std=[0.229, 0.224, 0.225]
        )
    ])
    return transform(img).unsqueeze(0)

def preprocess_keras(img, crop_name):
    img = img.convert("RGB")
    size = KERAS_INPUT_SIZES.get(crop_name, 224)
    img = img.resize((size, size))
    arr = np.array(img).astype("float32") / 255.0
    return np.expand_dims(arr, axis=0)

# ================= MODEL REGISTRIES =================

PYTORCH_MODELS = {}
KERAS_MODELS = {}
LABELS = {}

# ================= LOAD MODELS =================

def load_models():
    for file in os.listdir(MODELS_DIR):
        name, ext = os.path.splitext(file)
        crop_name = name.replace("_model", "").lower()

        model_path = os.path.join(MODELS_DIR, file)
        label_path = os.path.join(LABELS_DIR, f"{crop_name}_labels.json")

        if not os.path.exists(label_path):
            raise FileNotFoundError(f"Missing label file: {label_path}")

        with open(label_path, "r") as f:
            LABELS[crop_name] = json.load(f)

        # ---------- PyTorch ----------
        if ext == ".pth":
            num_classes = len(LABELS[crop_name])
            model = models.resnet18(weights=None)
            model.fc = torch.nn.Linear(model.fc.in_features, num_classes)
            model.load_state_dict(torch.load(model_path, map_location="cpu"))
            model.eval()
            PYTORCH_MODELS[crop_name] = model

        # ---------- Keras ----------
        elif ext in [".keras", ".h5"]:
            KERAS_MODELS[crop_name] = tf.keras.models.load_model(
                model_path,
                custom_objects={
                    "swish": tf.keras.activations.swish,
                    "FixedDropout": FixedDropout,
                    "EfficientNetB3": EfficientNetB3,
                },
                compile=False
            )

# Load models at startup
load_models()

# ================= PREDICTION =================

def predict(image, crop_name):
    crop_name = crop_name.strip().lower()

    if crop_name in PYTORCH_MODELS:
        model = PYTORCH_MODELS[crop_name]
        labels = LABELS[crop_name]
        tensor = preprocess_pytorch(image)
        with torch.no_grad():
            output = model(tensor)
            probs = torch.softmax(output[0], dim=0)
            idx = probs.argmax().item()
        return labels[idx], float(probs[idx])

    elif crop_name in KERAS_MODELS:
        model = KERAS_MODELS[crop_name]
        labels = LABELS[crop_name]
        arr = preprocess_keras(image, crop_name)
        preds = model.predict(arr, verbose=0)[0]
        idx = int(np.argmax(preds))
        return labels[idx], float(preds[idx])

    else:
        raise ValueError(f"No model found for crop: {crop_name}")