Quantitative comparison about vps HOT 7 CLOSED

HI @Scatteredrain

All amazing things are listed here: https://github.com/GewelsJI/VPS?tab=readme-ov-file#5-vps-benchmark

Thanks.

from vps.

@GewelsJI Thanks for your quick reply! But in this link: https://pan.baidu.com/s/1Qu-0l9w0ja92nzrlWMRhoQ
I can't find the prediction maps on the two Seen datasets of some models (only Unseen), like UNet and other image-based methods.
I wonder if you uploaded these results.

from vps.

We actually release all the pred results in the download link, and everything is available for the community. Could you please show me the file extracted from this tar file?

Best

from vps.

I think you may not give the results because the Image-based approach experiments on Seen datasets are not shown in your paper(PNS+)

from vps.

Sorry, I just mis-understood your question. We did not evaluate the performance of image-based methods on seen datasets as we are interested in the spatial-temporal association abilities during this stage. You can term seen datasets as a validation set, so we disregard them. More details can be found in our paper:

Hope you best!

from vps.

And as for the trained weights, I think it is not hard to implement (I mean UNet-family models) and to retrain them, you can try it then. Happy to see some interesting results from your side.

Hope our work helps you a lot. Thanks

from vps.

Thanks for the answer, that's exactly what I want to know!

from vps.