Open shuaiqikai opened 1 year ago
Theorectically it is generic for different usages, since the input of CAA is NxC while the output is also NxC (N can be a flaten form of WxH).
We have only tested it on poin cloud classification/segmentation/detection, as presented in the paper.
Is this CAA module only applicable to this paper? What about other methods that are not attention mechanisms?