fix_model.py

import tensorflow as tf
from tensorflow.keras import layers, models

# ======================
# REBUILD MODEL
# ======================
base_model = tf.keras.applications.MobileNetV2(
    input_shape=(224, 224, 3),
    include_top=False,
    weights="imagenet"
)

# Freeze same layers
for layer in base_model.layers[:-20]:
    layer.trainable = False

# IMPORTANT: No augmentation here (for inference)
model = models.Sequential([
    base_model,
    layers.GlobalAveragePooling2D(),
    layers.BatchNormalization(),
    layers.Dense(64, activation='relu'),
    layers.Dropout(0.5),
    layers.Dense(1, activation='sigmoid')
])

# ======================
# LOAD WEIGHTS ONLY
# ======================
model.load_weights("image_quality_model_final.h5")

# ======================
# SAVE CLEAN MODEL
# ======================
model.save("image_quality_model.keras")

print("✅ Model fixed and saved successfully!")