Susmit Agrawal

> Thanks for the PR. To prevent performance regression, it's better to do the following thing > > ```python > def _adaptive_pooling_1d(inputs, output_size, data_format, reduction_function): > def divisible_case(): > #...

Thank you for your quick response. I've downloaded the preprocessed dataset directly (lite version) using `./fetch_slip.sh preprocess`. I'm running `train_cnn3d.py` using the python command commented in line 10 of the...

This was done on purpose, to prevent unexpected behaviour for different input sizes, as well as offer flexibility for shape. The inputs can be clipped/padded as per requirement before passing...

The way PyTorch accomplishes this is well explained here: https://discuss.pytorch.org/t/what-is-adaptiveavgpool2d/26897 The issue is the rounding, instead of floor/ceil. This gives unequal size pieces that can't be sliced and stacked for...

I have created a PR with the updated version of Adaptive Pooling. However, this is an order of magnitude slower than the PyTorch version. Methods to improve performance need to...