sha2nkt / ipman-r Goto Github PK

Hi, thanks for sharing this amazing work!

When running the eval.py, I encounter an issue because I can't find a directory "ps".
I found that eval.py needs the images and models in "ps/scratch" and "ps/project".
So I am wondering that will files in "ps" be downloaded automatically? Or is it possible for me to download it manually?

Hi,

Are there any plans to publicly release a demo script, using which we can test the method on in-the-wild images

Hi, sorry for bothering you again.

Because Stability Loss and Ground Loss use 20000 vertices to do a series of calculations, it makes the training time much longer than the cases without using them, especially if hardware resources are limited.

Could you give me some advices to accelerate the calculation of the losses?
(Ex. vectorization of a function, parallelism techniques..etc.)

Hi, thanks for this inspiring paper!
I am curious about the scatter plot for distribution of poses on page 9 because I want to visualize my datasets, too.
Although I know it was produced by t-SNE, I still wonder how to process the poses with different frame numbers.

Hi, thanks for the amazing work!

I couldn't find code where the the sparse regressor for the sampled surface points is analytically computed. Is that available? As noted in the paper, I also checked the code base for the Self-contact paper and couldn't it their either.

Thanks again!

Hi, sorry for bothering you.
Whenever I want to train the dataset, after I execute the training code, in the near of line 250 of base_ dataset. py, there is always an error message indicating “local variable 'img' referenced before assignment”. I would like to ask how you can resolve this issue.

Hi, thanks for your great work! I'm wondering if you have released the evaluation code for MoYo. I found that the released model 2022_03_01-06_31_55_epoch1_stepcount_24000.pt can not perform well in the yoga pose.