Conv2d takes care of applying a convolutional filter on our MNIST images. Let's try to understand how convolution is applied on a one-dimensional array, and then move to how a two-dimensional convolution is applied to an image. We will look at the following image, to which we will apply a Conv1d of a filter (or kernel) size 3 to a tensor of length 7:
The bottom boxes represent our input tensor of seven values, and the connected boxes represent the output after we apply our convolution filter of size three. At the top-right corner of the image, the three boxes represent the weights and parameters of the Conv1d layer. The convolution ...
