Convolutional Neural Network visualization - weights or activations?

问题

Is the above visualization a rendering of the weights of the first convolutional layer or the activations on a given input image on the first convolutional layer?

Below is a visualization of the weights of the first convolutional layer of the Inception v2 model that I've been training for 48 hours:

I'm sure I have not converged my model after only 48 hours (on a CPU). Shouldn't those weights begin to smooth out by now, where training accuracy is over 90%?

回答1:

According to ImageNet Classification with Deep Convolutional Neural Networks, The first convolutional layer takes as input a 224x224x3 image (note: should be 227x227 instead of 224x224) and filters it with 96 kernels of size 11x11x3.

The figure above shows the convolutional kernels learned by the first convolutional layer.

note: in fact in the figure is possible count 96 convolutional kernels that sould be of size 11×11×3.

回答2:

In this particular The weight of the first layer of an AlexNet is being visualized. In terms of Activation visualization the visuals are not as smooth as visuals of weights.

Let me know if you have questions

来源：https://stackoverflow.com/questions/51616888/convolutional-neural-network-visualization-weights-or-activations