some problems about global context and cls_token

ma-xu / pointMLP-pytorch

[ICLR 2022 poster] Official PyTorch implementation of "Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework"

Apache License 2.0

501 stars 65 forks source link

some problems about global context and cls_token #66

Closed mmiku1 closed 2 years ago

mmiku1 commented 2 years ago

Hi @ma-xu gmp_list.append(F.adaptive_max_pool1d(self.gmp_map_list[i](x_list[i]), 1))

https://github.com/ma-xu/pointMLP-pytorch/blob/b5dcf4d0ca2085d10fe1dc5e6f2972cddec8399d/part_segmentation/model/pointMLP.py#L441

The features of each layer of the encoder are concat. After max pooling,the feature is concat with the last layer of the decoder. Why not concat with each layer of the decoder?

Thanks.

ma-xu commented 2 years ago

@mmiku1 It could be, but unnecessary. Empirically, this setting can achieve promising performance.

mmiku1 commented 2 years ago

Thank you for your answer. Wish you a happy life！