OpenGVLab / VideoMamba

VideoMamba: State Space Model for Efficient Video Understanding
https://arxiv.org/abs/2403.06977
Apache License 2.0
660 stars 47 forks source link

About the Mamba module #26

Closed qiuzidian closed 2 months ago

qiuzidian commented 2 months ago

image I would like to know if the mamba block in your work includes skip connection like vmamba, because I found in the code that the residual tensor seems to serve this. It seems that the figure of mamba block in your paper does not include it, was it my mistake?

Andy1621 commented 2 months ago

Yes, the residual is used undoubtedly. As for the Fig.2, I have claimed in the caption: We omit the initial normalization and the final residual for simplification.

qiuzidian commented 2 months ago

Thank you, I think I overlooked it