rose-stl-lab / equivariant-net Goto Github PK

Dear Repository Owner,

I hope this message finds you well. I've been working on reproducing the results for ResNet-rot on the RBC dataset based on your codebase, and I've run into some challenges.

The RMSE results I obtained were notably smaller than the values reported in your paper. Upon closer inspection, I identified two issues:

1. In the utils.py file, specifically in the test_epoch function (line 118), it seems that the RMSE calculation is missing the square root operation.
2. Additionally, I noticed a minor typo in data_prep.py, where the variable "img" should be replaced with "rot_img" on line 111.

Despite correcting these issues, the numerical results I obtained are still considerably smaller than those reported in the original paper. My results for Future RMSE and Domain RMSE of ResNet-rot on the RBC dataset are 0.15 and 0.16, respectively, while the values reported in Table 2 of your paper are 0.65 and 0.76.

I'm reaching out to seek guidance on how to obtain results consistent with those presented in your paper. Could there be differences in the data processing pipeline or measurement? Any insights or suggestions you could provide would be greatly appreciated. Thank you for your attention to this matter.

Your Pytorch version doesn't look right in README.md file.

pytorch 10.1 -> pytorch 1.01

I could not find the Rayleigh–Bénard convection DataSet. Is there a new repository from which I could retrieve this data set?
Thank you in advance.