Hi, I’m reading https://gluon-cv.mxnet.io/_modules/gluoncv/data/transforms/presets/ssd.html. I do not understand this part of the docstring: "If anchors is ``None``, the transformation will not generate training targets. Otherwise it will generate training targets to accelerate the training phase since we push some workload to CPU workers instead of GPUs." What are those training targets about?
I believe the targets here are just the bounding boxes for the training data with the transforms and augmentations applied to them.
My attempt at a guess is if ‘anchors’ is ‘None’ the transforms are not applied to the bounding boxes.