falaktheoptimist
commented
6 years ago Hi! If this is the output of the first layer, then there seems to be something wrong with the model itself - perhaps, training missing or its weights overshot. If this occured in the end layers, it would make sense, but not sure why the initial layers would produce such an output. Also, please make sure the image you passed in is not constant/ blank. That could be another reason which could cause this output. Thanks!
Hello.
I have visualized my work using your first method based on Visualizing and Understanding Convolutional Networks. However, I am facing the problem with the first convolution layers. They appear to be blank for all the feature maps.
I tried instead of using 55 conv layer, I used two 33 conv layers. The thing was second one followed by the first 3*3 conv layer appears fine but visualizing the first conv layer seems to encounter the problem.
How can i fix this problem? regards, Chanjong Im