neuro-ml / deep_pipe

A collection of tools for deep learning experiments
MIT License
37 stars 9 forks source link

PyramidPooling layer speed #89

Open AnihilatorGun opened 4 months ago

AnihilatorGun commented 4 months ago

level=0 and level=1 are too slow because of the peculiarity of the implementation. Mb start from the last level (level=self.levels - 1) and calculate subsequent levels using it?