vchoutas / smplify-x

Expressive Body Capture: 3D Hands, Face, and Body from a Single Image
https://smpl-x.is.tue.mpg.de/
Other
1.77k stars 340 forks source link

NaN loss value, stopping! #147

Closed man1989 closed 3 years ago

man1989 commented 3 years ago
python smplifyx/main.py --config cfg_files/fit_smplx.yaml --data_folder ./data 
--output_folder ./output --visualize False --model_folder ../smplx/models 
--vposer_ckpt ./vposer --part_segm_fn smplx_parts_segm.pkl

After running this i am getting below error

Found Trained Model: ./vposer/snapshots/TR00_E096.pt
init_t:  tensor([[0., 0., nan]], device='cuda:0')
NaN loss value, stopping!
Camera initialization done after 1.3177
Traceback (most recent call last):
  File "smplifyx/main.py", line 273, in <module>
    main(**args)
  File "smplifyx/main.py", line 262, in main
    **args)
  File "/mnt/hdd0/manishkr/smplify-x/smplifyx/fit_single_frame.py", line 367, in fit_single_frame
    cam_init_loss_val))
TypeError: unsupported format string passed to NoneType.__format__
(smplifyx)

Any help on this as why i am getting this error?

Update:- After disabling interactive mode and disabling cuda and interpenetration Mesh got generated but it has NaN co-ordinates

man1989 commented 3 years ago

Closing this as i was having setup issue

SoftwareDevJake commented 2 years ago

How was your setup problem? I'm having same issue, any detail solutions?

xiyichen commented 2 years ago

I had the same problem and turned out the NaN values came from the output from the decoder of VPoser. The issue was with the version of VPoser (as mentioned here: https://github.com/vchoutas/smplify-x/issues/144#issuecomment-854178835). You can try to install VPoser v1.0 from branch cvpr19 (pip install -e git+https://github.com/nghorbani/human_body_prior@cvpr19#egg=human_body_prior) and it should work.