will spark work well on depthwise conv?

keyu-tian / SparK

[ICLR'23 Spotlight🔥] The first successful BERT/MAE-style pretraining on any convolutional network; Pytorch impl. of "Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling"

https://arxiv.org/abs/2301.03580

MIT License

1.41k stars 82 forks source link

will spark work well on depthwise conv? #37

Closed LinB203 closed 1 year ago

LinB203 commented 1 year ago

hi, wonderful work! I wonder that will it work well while in_channel == group in nn.Conv2D?

keyu-tian commented 1 year ago

Yeah for sure