[Pooling] Stride default size issue
DonghakPark opened this issue · comments
Issue
- If the user does not enter the stride value when creating the pooling2d layer in the current NNTrainer, 1 is set to the default value.
- in other framework (tensorflow, pytorch) they set default value as kernel size
- default stride size = kernel size is reasonable in most case -> user use pooling for reduce size
NYI) below is other framework's MaxPool2D's args
tensorflow : https://www.tensorflow.org/api_docs/python/tf/keras/layers/MaxPool2D#args
pytorch : https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html