Implementation of convolutional layers.
I wrote this convolutional layer and convolutional network as an introduction to computer vision and applying deep learning to process and analyze graphical data.
The layer I created is similar to the torch.nn.Conv2d, as shown in the ConvoNet file. It supports dilation, strides and same padding.
Doing it for RGB images was challenging, and I ended up learning a lot about how to manipulate tensors and how images are processed in convolutional layers.
 |
 |
 |
 |
 |
 |
| No padding, no strides | Arbitrary padding, no strides | Same padding, no strides | Full padding, no strides | Padding, strides | Dilation |
I copied the gifs from this repo.