Visualizing learned features of a caffe neural network
After training a convolutional neural network, one often wants to see what the network has learned. The following python function creates and displays an image with all convolutions of a specific layer as shown above. If you supply the function with a filename the image will also be saved on disk.
import numpy as np import matplotlib.pyplot as plt def visualize_weights(net, layer_name, padding=4, filename=''): # The parameters are a list of [weights, biases] data = np.copy(net.params[layer_name].data) # N is the total number of convolutions N = data.shape*data.shape # Ensure the resulting image is square filters_per_row = int(np.ceil(np.sqrt(N))) # Assume the filters are square filter_size = data.shape # Size of the result image including padding result_size = filters_per_row*(filter_size + padding) - padding # Initialize result image to all zeros result = np.zeros((result_size, result_size)) # Tile the filters into the result image filter_x = 0 filter_y = 0 for n in range(data.shape): for c in range(data.shape): if filter_x == filters_per_row: filter_y += 1 filter_x = 0 for i in range(filter_size): for j in range(filter_size): result[filter_y*(filter_size + padding) + i, filter_x*(filter_size + padding) + j] = data[n, c, i, j] filter_x += 1 # Normalize image to 0-1 min = result.min() max = result.max() result = (result - min) / (max - min) # Plot figure plt.figure(figsize=(10, 10)) plt.axis('off') plt.imshow(result, cmap='gray', interpolation='nearest') # Save plot if filename is set if filename != '': plt.savefig(filename, bbox_inches='tight', pad_inches=0) plt.show()
The following is an example of how to use this function
from visualize_caffe import * import sys # Make sure caffe can be found sys.path.append('../caffe/python/') import caffe # Load model net = caffe.Net('/home/smistad/vessel_net/deploy.prototxt', '/home/smistad/vessel_net/snapshot_iter_3800.caffemodel', caffe.TEST) visualize_weights(net, 'conv1', filename='conv1.png') visualize_weights(net, 'conv2', filename='conv2.png')
As usual the code can be found on my GitHub page