search

D-X-Y / AutoDL-Projects

Automated deep learning algorithms implemented in PyTorch.
MIT License
1.56k stars 281 forks source link

Concern about CWI (channel-wise interpolation) in TAS. #34

Closed Fangyu-ict closed 4 years ago

Fangyu-ict commented 4 years ago

Hi, thanks for your great work.

After reading your paper TAS, I don't understand the channel-wise interpolation. Could you explain or give an insight on channel-wise interpolation (i.e., making weighted sum on feature maps of different sizes), and the corresponding effects.

Thanks.

D-X-Y commented 4 years ago

Thanks for your interest. The CWI operation aims to align the shape of different feature maps. For example, to aggregate two feature maps A in R^{HWC1} and B^{HWC2}, you can not weighted-sum them due to the inconsistent shape. So we use CWI (https://github.com/D-X-Y/NAS-Projects/blob/master/lib/models/shape_searchs/SoftSelect.py#L41) to align their shapes.

Fangyu-ict commented 4 years ago

Thanks.