src/siamese_model.py

#src/siamese_model.py
import tensorflow as tf
from tensorflow.keras import layers, Model
from tensorflow.keras.utils import plot_model

IMG_SIZE = (128, 128, 1)

def build_embedding():
    inp = layers.Input(shape=IMG_SIZE)
    x = layers.Conv2D(32, 3, activation="relu")(inp)
    x = layers.MaxPooling2D()(x)
    x = layers.Conv2D(64, 3, activation="relu")(x)
    x = layers.MaxPooling2D()(x)
    x = layers.Flatten()(x)
    x = layers.Dense(128, activation="relu")(x)
    return Model(inp, x, name="embedding")

def build_siamese_model(visualize=False):
    embedding = build_embedding()

    img1 = layers.Input(shape=IMG_SIZE, name="img1")
    img2 = layers.Input(shape=IMG_SIZE, name="img2")

    f1 = embedding(img1)
    f2 = embedding(img2)

    # Simplified absolute difference layer
    distance = layers.Lambda(lambda tensors: tf.abs(tensors[0] - tensors[1]))([f1, f2])
    output = layers.Dense(1, activation="sigmoid")(distance)

    model = Model(inputs=[img1, img2], outputs=output, name="Siamese_Network")
    
    if visualize:
        plot_model(model, to_file='siamese_architecture.png', show_shapes=True, show_layer_names=True)
        print("✅ Model architecture saved as 'siamese_architecture.png'")
        
    return model
    
# import tensorflow as tf
# from tensorflow.keras import layers, Model

# IMG_SIZE = (128, 128, 1)

# def build_embedding():
#     inp = layers.Input(shape=IMG_SIZE)

#     x = layers.Conv2D(32, 3, activation="relu")(inp)
#     x = layers.MaxPooling2D()(x)
#     x = layers.Conv2D(64, 3, activation="relu")(x)
#     x = layers.MaxPooling2D()(x)
#     x = layers.Flatten()(x)
#     x = layers.Dense(128, activation="relu")(x)

#     return Model(inp, x, name="embedding")

# def build_siamese_model():
#     embedding = build_embedding()

#     img1 = layers.Input(shape=IMG_SIZE, name="img1")
#     img2 = layers.Input(shape=IMG_SIZE, name="img2")

#     f1 = embedding(img1)
#     f2 = embedding(img2)

#     # ✅ SAFE, SERIALIZABLE (NO Lambda)
#     distance = layers.Subtract()([f1, f2])
#     distance = layers.Lambda(lambda x: tf.abs(x))(distance)

#     output = layers.Dense(1, activation="sigmoid")(distance)

#     return Model(inputs=[img1, img2], outputs=output)