search

alexhe101 / Pan-Mamba

Pan-Mamba: Effective Pan-Sharpening with State Space Model
https://arxiv.org/abs/2402.12192
80 stars 7 forks source link

Running time？ #5

Closed EricPaul03 closed 7 months ago

EricPaul03 commented 7 months ago

Hello, I've tried your mamba module in my project, but I found it is quite slower than self-attention, is it due to my sequence length(about 400) ?

alexhe101 commented 7 months ago

In our paper, as can be seen in Fig.5, when the resolution is set to 128*128 - equating to a sequence length of 16384, the run-time performance of our Mamba block is comparable to that of self-attention. Regarding your question, you could use the default Mamba block by changing "v3" to "v1" to ascertain if our modifications were the cause of this issue.

EricPaul03 commented 7 months ago

In our paper, as can be seen in Fig.5, when the resolution is set to 128*128 - equating to a sequence length of 16384, the run-time performance of our Mamba block is comparable to that of self-attention. Regarding your question, you could use the default Mamba block by changing "v3" to "v1" to ascertain if our modifications were the cause of this issue.

Thank you so much, I will try it.