search

wurenkai / UltraLight-VM-UNet

[arXiv] The official code for "UltraLight VM-UNet: Parallel Vision Mamba Significantly Reduces Parameters for Skin Lesion Segmentation".
232 stars 33 forks source link

Confusion about the PVM Layer #59

Open TaoZhou1206 opened 2 weeks ago

TaoZhou1206 commented 2 weeks ago

Hello, author! I see in the description of the PVM layer section in the latest edition of your paper: "Vision Mamba consists of Mamba with residual connections and adjustment factors "Is this Vision Mamba self-named? Or the visual mamba that was proposed some time ago? Is this a thing? I don't understand this part; Does the mamba in Figure 3 (a) refer to the most basic mamba model? Because I have read your last edition of the paper, it was originally written VSS block, but it was changed to Mamba in the new edition, so I don't quite understand what is used, and I am afraid of misunderstanding! I really hope you can answer my questions! Thank you very much!

wurenkai commented 2 weeks ago

Hi, the PVM layer is depicted in the latest version of Figure 3a and is composed of basic mamba combined with residual connections and adjustment factors. If still having problems, we suggest that reading our work together with the code may be more helpful for you.

TaoZhou1206 commented 2 weeks ago

Hello, author! I have read your code, and the Mamba in PVM Layer in the code is from "mamba_ssm import Mamba", so the Mamba in PVM Layer is the basic Mamba model used, which has nothing to do with visual mamba, is that what it means? (Because I have been reading your previous version of the paper and code some time ago, thinking that the PVM Layer is used in the VSS block, based on the VSS block to expand and improve) Thank you very much for your answer!