Newbie question concerning the time required to categorise images into one of n classes (in my case different animals): if my model categorises an image into 2, 3, 4, 5… classes of animal, is the time taken in the same ratio (2:3:4:5…) or does it scale differently?
Also, how are the memory requirements likely to scale?
Hi @acutetech this will have no effect, as the number of classes only affects the output layer (for normal vision models), or has no effect on the output layer for object detection models (as this is determined by the number of bounding boxes / labels you want to find). Naturally there will be effects on accuracy when going to smaller models - but not a big concern with the current object detection transfer learning model.