The image dataset undergoes some pre-processing to become a column vector so that it can be inputted. Why are they not processed into a row vector and inputted instead?
Either row or column vectors may be used. There is little standardization in how the data set is organized.
2 Likes
Sometimes while using the lib we have to follow its way of using and accepting things, ways in which a particular lib expects the data to be, so its better to make it a habit from the start to deal with the data in that format.
For example, TensorFlow typically uses channel-last format (height, width, channels), while PyTorch often uses channel-first format (channels, height, width).
When flattening, these conventions can influence whether you choose row-wise or column-wise flattening.
like h x w x c is 4 x 4x 3 becomes 16 x 3 (column wise)
and c x h x w is 3 x 4 x 4 becomes 3 x 16 (row wise)
2 Likes