Keras-模型中间层特征图可视化

CNN特征图可视化

实验过程需要观察模型中间层的输出，通过特征图观察当前层是否作用。本代码环境为tensorflow+keras.

# Define model
encoder_model, autoencoder_model = make_model(input_S.shape[1:],"identity")
# Load model
autoencoder_model.load_weights('model1013-1.hdf5')

# Input test images
test_S = X_test[0]
test_S = np.reshape(test_S,[1,263,263,4])
test_C = X_test[1000] 
test_C = np.reshape(test_C,[1,263,263,4])

output_C = Model(inputs=encoder_model.get_input_at(0),outputs=encoder_model.get_layer(index=31).output)
output_C_output = output_C.predict([test_S,test_C])
# index=31: 指定输出层

def get_row_col(num_pic):
    col = 10
    row = num_pic//col if num_pic > 5 else 1
    
    return row, col
 
def visualize_feature_map(img_batch):
    feature_map = np.squeeze(img_batch, axis=0)
    print("shape:",feature_map.shape)
 
    feature_map_combination = []
    plt.figure()
 
    num_pic = feature_map.shape[2]
    row, col = get_row_col(num_pic)
 
    for i in range(0, num_pic):
        feature_map_split = feature_map[:, :, i]
        feature_map_combination.append(feature_map_split)
        plt.subplot(row, col, i + 1)
        plt.imshow(feature_map_split)
        plt.axis('off')

    plt.savefig('output_C.jpg')
    plt.show()
    
# show the feature map
visualize_feature_map(output_C_output)

Example: feature map

参考

1: Visualizing what convnets learn.
2: Keras可视化遥感图像.