Neuman dataset
Opened this issue · 9 comments
Thank you for your code!!
I tried your code on neuman dataset. but result was bad,
It seems preprocess work well, can you help us for debugging?
I shared preprocessed dataset here
https://drive.google.com/file/d/1ZQ8Wwz_3DmCte07zpoQKHxFot5zEAXuJ/view?usp=drive_link
did you change some training code?
No I didn't.
I used released colmap data from neuman, and work preprocess code.
then tried training code in Neuman mode.
since smplx_params_smoothed did not work, we used smpl_optimized/smplx_params
The original code does not use smplx_params_smoothed for NeuMan dataset, while uses it for Custom dataset. Why did you try smoothed one? Did you run train with setting dataset='NeuMan' in here(
ExAvatar_RELEASE/avatar/main/config.py
Line 40 in 69e12bb
Yes, I did.
I run training with setting dataset='NeuMan'
I haven't seen these kinds of failure cases... could you share your rendered images on NeuMan dataset?
Do you mean the rendering result?
here is the result of main/test.py
https://drive.google.com/file/d/1-DfQ-hG6MLiQU97YXAbJgv03L37BPVA7/view?usp=drive_link
It seems you're loading some wrong data from the dataloader like wrong camera parameters or wrong smplx parameters. could you check again? please try with newly cloned without any modification.
I recreated dataset, and it was because of data index problem.
after recreation, training worked well! thank you.