Well, when I finally upgraded my final project codes, I found a very interesting thing that I tried to reduce the feature dimensions after each conv layers. I had 3 conv layers and the input image have 3 dimensions and 28X28 size. The size would not change because I used padding to keep size the same. After the first conv layer the dimension became 16. And after the last layer the feature dimensions became 128. However, I wanted to use pooling layer to reduce the feature dimensions. I want to reduce it to 16 so that I can connect another layer after it without calculating the dimensions or input channels. Because if I do not do that the dimension might increase to 1000 or more.
Do you mean the number of channels by “feature dimension”?
If so, you could reduce it by using another conv layer with the desired number of output channels.
If you want to decrease the image resolution, you could add a pooling layer after the conv layer.
If you want to reduce the number of feature channels, you could add a 1x1 conv layer after the conv layer without reducing the image resolution.
If you want to reduce both, you can add another one conv layer.
Yeah, that is what I mean. Well, I finally found a paper about this. They were using a 1X1 conv to reduce the out channels.
Anyway, thank you.
Yeah, I tried to reduce the output channel and another paper said the same methods.
But anyway, thank you so much.
I am having the same (similar) problem here. Any ideas how to solve it?