CODEWITHSUNDEEP
pythonkerasdeep-learningconv-neural-networktensorflow2.0
cUser
cUser

Reputation: 469

How can i get the all output keras layers (features maps) during prediction phase?

I tried to take cfiar data set and to try to get the feature map of every output giving as an input one of the test images.

(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()

# Normalize pixel values to be between 0 and 1
train_images, test_images = train_images / 255.0, test_images / 255.0

class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer',
           'dog', 'frog', 'horse', 'ship', 'truck']

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10))

model.compile(optimizer='adam',
          loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
          metrics=['accuracy'])

history = model.fit(train_images, train_labels, epochs=10, 
                validation_data=(test_images, test_labels))

test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)

layer_input = test_images[0]                  

for i in range(len(model.layers)):
    get_layer_output = K.function(inputs = model.layers[i].input, outputs = model.layers[i].output)
    print(get_layer_output(layer_input))
    layer_input = model.layers[i].output

My feeling is that i misunderstand something regarding how to set the input and also how to take the output during the prediction.

Upvotes: 2

Views: 948

Answers (2)

Nuwan madhusanka
Nuwan madhusanka

Reputation: 111

You can use following python script to draw feature maps of the neural network.

import tensorflow as tf
import  cv2
import numpy as np

# use 'model' as your model name
# change 224 image size to your model input image size 

layer_names = [layer.name for layer in model.layers]

layer_outputs = [layer.output for layer in model.layers[1:]]

feature_map_model = tf.keras.models.Model(inputs=model.input, outputs=layer_outputs)

img = cv2.imread("path/to/image/image.jpg")
img = cv2.resize(img, (224,224))
img = np.array([img])/255.0

feature_maps = feature_map_model.predict(img, verbose=0)

for layer_name, feature_map in zip(layer_names, feature_maps):

    if len(feature_map.shape) == 4:
    k = feature_map.shape[-1]
    size=feature_map.shape[1]

    image_belt = np.zeros((size, size * k))

    for i in range(k):
        feature_image = feature_map[0, :, :, i]
        feature_image-= feature_image.mean()
        feature_image/= feature_image.std ()
        feature_image*=  64
        feature_image+= 128
        feature_image= np.clip(feature_image, 0, 255).astype('uint8')
        image_belt[:, i * size : (i + 1) * size] = feature_image

    scale = 20. / k
    plt.figure( figsize=(scale * k, scale) )
    plt.title ( layer_name )
    plt.grid  ( False )
    plt.imshow( image_belt, aspect='auto')

Reference : https://www.analyticsvidhya.com/blog/2020/11/tutorial-how-to-visualize-feature-maps-directly-from-cnn-layers/

Upvotes: 1

TF_Support
TF_Support

Reputation: 1836

You should set the input to model.layers[0].input if you are using Sequential model.


First, expand the dimensionality of the Test Input to include the Batch_Size:

layer_input = test_images[0]    
plt.imshow(layer_input)   # Plot Test Image
layer_input = tf.expand_dims(layer_input,0)   # Add prefix of Batch Size 
print(layer_input.shape)  # Prints  : (1, 32, 32, 3)

Output:

original_image

Modified code for plotting:

for i in range(len(model.layers)):
    get_layer_output = K.function(inputs = model.layers[0].input, outputs = model.layers[i].output)

    get_1_output = get_layer_output(layer_input)
    # print(get_1_output.shape) << Use this to check if the Output shape matches the shape of Model.summary()

    if get_1_output.ndim == 4:             # Check for Dimensionality to plot ONE feature map (Batch size, Length, Width
      plt.imshow(get_1_output[0][:,:,:3])  # Plots the output of Conv2D and MaxPooling
    else:
      print(get_1_output)   # If not Image, ie. Array, print the Values

    plt.show()

Output:

feature_maps


I hope I answered your question.

Upvotes: 2

Related Questions