similarity_detection.cpython-312.pyc
1 � 2 � �e � �b � d dl mZmZmZ d dlZd dlmZ d dlmZ d dl m 3 Z 4 d� Z d 5 d�Z d 6 d�Z d 7 d �Zy) � )�branches�branches_nopool� 8 branch_cvaN)�keras)�layers)�l2c � � | \ }}t j j t j j ||� � }|S )N)�tf�math�abs�subtract)�vects�x�y�results �\C:\Users/dvalsamis/change/Change_detection_SSL_Siamese\architectures\similarity_detection.py�abs_diffr s3 � � 9 �C�A�a� �W�W�[�[����)�)�!�A�.� /�F� �M� c � � t j t | � t |� t |� f� }t j t | � t |� t |� f� }t | ||� } ||� } ||� } t j t 10 � ||g� } t j ddddd�� |� } t j ddd � 11 � | � } 12 t j d d � � | 13 � } t j � | � } t j ddd�� | � } t j ddd�� | � } t j ddd�� |� }t j ||g|d�� }|S )N� �� r �relu�same�joint)� 14 activation�padding�nameg�������?� � dropout_joint��seedr )� r# � 15 pool_joint)� pool_sizer � �dense1�r r �@ �dense2�sigmoid�output� pretext_task1��inputs�outputsr ) r �Input�intr �Lambdar �Conv2D�Dropout� MaxPooling2D�Flatten�Denser �Model)� 16 IMG_HEIGHT� IMG_WIDTH� IMG_CHANNELS�input_1�input_2�embedding_network�x1�x2� merge_layerr �dr_jointr$ �flattenr' r* � output_layer�siameses r �pretext_task_onerG sR � ��l�l�C� 17 �O�S��^�S� �=N�O�P�G��l�l�C� 18 �O�S��^�S� �=N�O�P�G�!�*�i� �F�� �7� #�B� �7� #�B� *�&�-�-��)�2�b�'�2�K� W�F�M�M�"�f���g� V�Wb� c�E�A�f�n�n�S�q� �A�%�H�H�K��$�$�v�l�K�H�U�J� �f�n�n� �z�*�G� @�V�\�\�#�&�x� @�� I�F� ?�V�\�\�"��h� ?�� G�F�G�6�<�<��i�h�G��O�L��k�k�'�7�!3�\�P_�`�G� �Nr c 19 � � t j t |� t |� t |� f� }t j t |� t |� t |� f� }t | ||||� } ||� } ||� } t j t 20 � || g� } 21 t j dddt |� t |� dd�� | 22 � } t j | dd� � | � } t j � | � } t j d 23 dd � � | � } t j d dd� � |� } t j ddd� � |� }t j ||g|d�� }|S )Nr r r r r )r �kernel_regularizer�bias_regularizerr r r r r! r& r' r( r) r* r+ r, r- r. ) r r1 r2 r r3 r r4 r r5 r7 r8 r r9 )�dropout�decayr: r; r< r= r> r? r@ rA rB r rC rD r' r* rE rF s r �pretext_task_one_nopoolrM & sc � ��l�l�C� 24 �O�S��^�S� �=N�O�P�G��l�l�C� 25 �O�S��^�S� �=N�O�P�G�(��%��Y�P\�]�� �7� #�B� �7� #�B� *�&�-�-��)�2�b�'�2�K� Q�F�M�M�"�f��B�u�I�hj�kp�hq� |B� IP� Q� R]� ^�E�E�f�n�n�W�1�_�E�e�L�H� �f�n�n� �x�(�G� @�V�\�\�#�&�x� @�� I�F� ?�V�\�\�"��h� ?�� G�F�G�6�<�<��i�h�G��O�L��k�k�'�7�!3�\�P_�`�G� �Nr c � � t j t |� t |� t |� f� }t j t |� t |� t |� f� }t | |d|||� } ||� } ||� } t j t 26 � || g� } 27 t j ddt |� t |� dd�� | 28 � } t j d�� | � } t j d d 29 �� | � } t j | d d � � | � } t j � |� } t j dd d�� |� } t j dd d�� |� } t j d dd�� |� }t j ||g|d�� }|S )Nr r r r r )rI rJ r r � 30 norm_joint)r r � 31 relu_jointr r r! r& r' r( r) r* r+ r, r- r. )r r1 r2 r r3 r r4 r �BatchNormalization� 32 Activationr5 r7 r8 r r9 )rK rL r: r; r< r= r> r? r@ rA rB r �batch_norm_joint�activation_joint� 33 drop_jointrD r' r* rE rF s r � pretext_onerV @ s� � ��l�l�C� 34 �O�S��^�S� �=N�O�P�G��l�l�C� 35 �O�S��^�S� �=N�O�P�G�#�G�U�1�Z��L�Y�� �7� #�B� �7� #�B� *�&�-�-��)�2�b�'�2�K� ~�F�M�M�"�f��E��UW�X]�U^�hn�u|� }� J� K�E�C�v�0�0�l�C�E�J��C�v�(�(��l�C�DT�U��F�� � ��a�o�F�GW�X�J� �f�n�n� �z�*�G� @�V�\�\�#�&�x� @�� I�F� ?�V�\�\�"��h� ?�� G�F�G�6�<�<��i�h�G��O�L��k�k�'�7�!3�\�P_�`�G� �Nr )�` rW r )�architectures.branchr r r � 36 tensorflowr 37 r �tensorflow.kerasr �keras.regularizersr r rG rM rV � r r �<module>r] s+ �� F� F� � � #� !�� 38 �4�4r