Thank you for your nice work!! Question about evaluation code!!

[justinday123]

Hi. Thank you for your nice work!

The code at ImageClassification/demo_cls_hypersigma.ipynb is evaluation code in your paper? Thanks!

Have a nice day!

[jinyaoWHU]

@justinday123 Hi, thank you for your kind words!
Actually, we used the code in ImageClassification/demo_seg_hypersigma.ipynb for evaluation in our paper.

[justinday123]

Thanks a lot for your response!!
Also, could you please share the benchmark results for ViT-Base? If it’s difficult to provide all of them, even a partial summary would be greatly appreciated.
Also, may I ask why experiments weren’t conducted with ViT-Small?

Thank you once again, and I hope everything goes well for you!

[justinday123]

Also, I understand that linear probing usually involves freezing the backbone and training only the linear layer, but it seems that the evaluation code does not follow this approach. Is this a unique evaluation method used in the HSI foundation work?
Thank you!

[jinyaoWHU]

@justinday123 Hello, thank you very much for your interest!
All the experimental results in our paper are based on the Base model. Currently, our model has only three versions — Base, Large, and Huge — and does not include a Small version.
In our experiments, we adopted a full fine-tuning strategy; however, in hyperspectral foundation model research, linear probing is also an optional evaluation approach.

[justinday123]

Thank you for your response!🙇

Then, when performing linear probing, the evaluation code removes parts of the pt file such as the patch embedding.
In that case, should I just use it without removing those parts?
When I remove, size error occurred, when I changed the patch size.

Thanks again! Have a nice day!