Open gexinyuan1997 opened 1 month ago
One more question. Why was convolution used in the shallow layer instead of PVM?
You can follow the split merge method of the PVM layer by splitting the channel in front of each HSSBlock or VSSBlock and then doing a concat merge later. This can be applied to any Mamba module. Shallow layers use convolution to increase the number of channels first so that the PVM Layer can be used later to split the channels correctly.
Hello! Thank you for your excellent work. I am very interested in Parallel VM UNET and Parallel H-VMUNET as mentioned in your paper. Can you share the related code?