Bad result on segmentation task

[SulaimanLi]

Hi, thank you for your excellent work which inspired me a lot.

I did some experiments on segmentation task by mmsegmentation toolbox (which version adaptation is really makes me insane XD:). However, I found that the results were not as good as other models of similar magnitude and calculation. But I do these on my private dataset which may act different from those famous and pulic datasets, so I'm not sure which brings me the performance gap, as the custom backbones doing good on classification task are also good at segmentation task.

I'm wondering that is the MambaVision suitable for downstream tasks? Could you provide me with more information to support me to keep trying. it's really confusing me, I will be very appreciate if you could reveal in advance whether the relevant experiments are going well.

[ahatamiz]

Hi @SulaimanLi

Thanks for the query. We have tested the model on ADE20K segmentation tasks and results are as reported in the paper. It would be hard to speculate whether we always outperform other models in all segmentation tasks, but our experimental results on ADE20K were promising. We will release the downstream tasks code including segmentation soon, in case it might be helpful.

[ahatamiz]

Code and models for semantic segmentation have been added.