fernandoperlar/preprocessing_image
0
1U
2 �T�a� � @ s� d dl mZ d dlmZ d dlmZ d dlZd dl Z3d dl Z d dl
mZmZmZmZmZmZmZ d dlmZ d dlmZ ddlT G dd � d �ZdS )4� )�FilterN)� callbacks�models�layers�utils�losses�
preprocessing�backend)�metrics)�5optimizers� )�*c @ s\ e Zd Zddd�Zdd� Zdd� Zddd �Zdd d �Zdd
d�Zdd� Z dd� Z6dd� Z d7S )�ModelTc C s� || _ || _| �� | _d| j � d�d d | _d| _t�| j |� tj j8| jd| j � d�ddd� |rp| j� � |r�t �
t �d| j � d��� t �d � t �� d S )9N� ./output/z /weights_z {epoch:03d}z.hdf5r z10/model.pngT)�to_file� show_shapes�show_layer_names�off)�name�filter�_Model__create_model�model� weights_path�11best_epoch�computer�create_output_folderr � vis_utils�12plot_model�summary�plt�imshow�imread�axis�show)�selfr r �newr �plot� r' �=c:\Users\ferna\Desktop\mammo-echo-thermo\src\scripts\Model.py�__init__
s 13 1415zModel.__init__c s� G � fdd�dt j�� t�� }|�t jdd�� |�� | jd�� |�t jdddd �� |�t jd16d �� |�t jdddd �� |�t � � � |�t j17d d
�� |�t j ddd �� |�t j18d d
�� |�t j ddd �� |S )Nc s2 e Zd Z�� fdd�Zdd� Z� fdd�Z� ZS )z)Model.__create_model.<locals>.FilterLayerc s$ || _ t� | �jf ddi|�� d S )Nr Z filter_layer)r �superr) )r$ r �kwargs)� FilterLayer� __class__r' r( r) # s z2Model.__create_model.<locals>.FilterLayer.__init__c S s6 |j }t�| j|gtjg�\}t�|�}|�|� |S �N)�shape�tf� py_functionr �float32r �
stop_gradient� set_shape)r$ �imager/ r' r' r( �call( s19 2021z.Model.__create_model.<locals>.FilterLayer.callc s22 t � �� S r. )r* �23get_config)r$ �r- r' r( r7 0 s z4Model.__create_model.<locals>.FilterLayer.get_config)�__name__�24__module__� __qualname__r) r6 r7 �
__classcell__r' �r, r8 r( r, " s r, )�� i � )r/ )r � )r? r? �relu)�25activation)� rC )� pool_size皙�����?)�raterC �softmax) r �Layerr �26Sequential�add�Inputr �Conv2D� MaxPooling2D�GlobalAveragePooling2D�Dropout�Dense)r$ r r' r= r( Z__create_model! s zModel.__create_modelc C sR | j jtjtjdd�dgd� td| j� d�d��}|�| j � � � W 5 Q R X d S )Ng{�G�zt?)�
learning_rate�accuracy)�loss� optimizerr27 r z /model.json�w)28r �compiler �binary_crossentropyr �Adam�openr �write�to_json)r$ Z json_filer' r' r( rV E s z
Model.compilec 29 s G � fdd�dt j�}t j| jd|r(dndddd�}|| j| j�}| jj|� |rTdnd|||gd �} |j| _|�rtj �30d31� t� � tj | j
d d d
� tj | j
d dd
� tj | j
d dd
� tj | j
d dd
� t�d� t�d� t�d� tjdd� t�� d S )Nc s" e Zd Zdd� Zd� fdd� ZdS )z Model.fit.<locals>.GetProgressc S s || _ || _d| _d| _d S )Nr )r r �
best_accuracyr )r$ r r r' r' r( r) M s z'Model.fit.<locals>.GetProgress.__init__Nc s� |d | j kr |d | _ || _td| j� d�d|d � d� � d� dt|d d�� d� | jj| jd d �� d32d � |d � kr�td d33d � d S )
N� val_accuracyz
Model z -> zEpoch r �/zAccuracy (Validation): rC ��epoch� )�end�34)r\ r �printr �roundr �format)r$ r` �logs��epochsr' r( � on_epoch_endS s 35 ���� z+Model.fit.<locals>.GetProgress.on_epoch_end)N)r9 r: r; r) rj r' rh r' r( � GetProgressL s rk r] r r T�max)�monitor�verbose�save_best_only�mode)ri rn �validation_datar ZggplotrS z
Training loss)�label�val_losszValidation lossrR zTraining accuracyzValidation accuracyzTraining Loss and AccuracyzEpoch #z
Loss/Accuracyz36lower left)�loc)r �Callback�ModelCheckpointr r r �fitr r �style�use�figurer& �history�title�xlabel�ylabel�legendr# )37r$ �train_generator�validation_generatorri rn r& rk �38checkpointZ get_progressr{ r' rh r( rw K s> 39�40�� 414243 z Model.fitNc C s, t �|d kr | jj| jd d�n|�| _d S )Nr r_ )r �44load_modelr rf r r )r$ �pathr' r' r( r� � s zModel.load_modelc
C s8 |d k r| � |� tj| j�|�dd�}t�|j|�}tj|t� |j45�d�j t j
jdd� t �� tt�|j|�� |d d }|d d }|d d }|d d }t|| �t|| | | � } tdt| d�� |t|| � }46td t|47d�� |t|| � } td48t| d�� |t|| � } td t| d�� d S ) N������r" )�confusion_matrix�display_labelsr )�cmap�xticks_rotationr z Accuracy:� z Specificity:z Sensitivity:z49Precision:)r� �np�argmaxr �predictr50 r� �classes�ConfusionMatrixDisplay�unique�labelsr& r �cm�Bluesr# rd �classification_report�floatre )
r$ r� r� � predictionsr� ZTPZTNZFPZFNrR � specificity� sensitivity� precisionr' r' r( �evaluate� s& 51$ zModel.evaluatec C s� t �t �|j�t j��d�d d }tj|dd�}| �|�}| � |j|�\}}t52j dddd �\}}|d � |d � |d �
tj�|j�� |d � tj�|�� |d �
d53|j� �� |d � |� |d �
d t�| j�|�d �� �� t54�� d S )
Nr2 g �?�� r r� r r? )� � )�figsizez Real class: rC zPredicted class: )�cv�cvtColorr! r5 �
COLOR_BGR2RGB�astyper� � expand_dimsZ_Model__compute_heatmapZ_Model__get_heatmapr �subplotsr � set_title�osr� �basenamer � array_to_img�categoryr� r r� r# )r$ r5 �img�heatmap� jet_heatmap�superimposed_img�fig�axr' r' r( �visualize_heatmap� s $55&zModel.visualize_heatmapc C s� | j jdd�}tj| j jg|j| j jgd�}t�� �8}||�\}}|�|d d �df |�}tj |dd�}W 5 Q R X |d |dtj56f }t� |�}t� |d�tj
�|� }|�� S ) NrC )�index)�inputs�outputsr )r r rC r� r .)r � get_layerr r �input�outputr0 � GradientTape�gradient� reduce_mean�newaxis�squeeze�maximum�math�57reduce_max�numpy) r$ r5 Z58last_layerZ59grad_model�tapeZ model_output�gradsZ pooled_gradsr� r' r' r( �compute_heatmap� s 60 61zModel.compute_heatmapc C s� t jj|dd�}t j�|�}t�d�}|t�d��d d �d d�f }|t�d| � }t j� |�}|�62|j d |j d f�}t j�|�}|d | }t j� |�}||fS )63N� grayscale)�64color_mode�jet� r? r� r r rE ) r r5 �load_img� img_to_arrayr� �get_cmapr� �arange�uint8r� �resizer/ )r$ r� r� r� r� Z65jet_colorsr� r� r' r' r( � get_heatmap� s 66 zModel.get_heatmap)TTT)TT)N)N) r9 r: r; r) r rV rw r� r� r� r� r� r' r' r' r( r s 67$68?6970 r )�loggingr �matplotlib.pyplot�pyplotr Z
matplotlib.cmr� r� r� �cv2r� �71tensorflowr0 �kerasr r r r r r r �sklearnr72 Ztensorflow.kerasr �Miscr r' r' r' r( �<module> s $