search

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

Segmentation fault #83

Closed zorropop closed 4 years ago

zorropop commented 4 years ago

Testing sample images found in smplify-x/src/homogenous-master/samples. Every time there is a segmentation fault at beginning of Stage 3:

~/smplify-x$ python3 smplifyx/main.py --config cfg_files/fit_smplx.yaml --data_folder data --output_folder output --visualize="True" --model_folder ../smpl-x/models --vposer_ckpt ../smpl-x/vposer_v1_0

Processing: data/images/im1097.jpg Found Trained Model: ../smpl-x/vposer_v1_0/snapshots/TR00_E096.pt Camera initialization done after 7.6183 Camera initialization final loss 1101.1987 Stage 000 done after 28.3231 seconds
Stage 001 done after 5.3000 seconds
Stage 002 done after 13.7803 seconds
Orientation: 0%| | 0/1 [00:47<?, ?it/s Segmentation fault (core dumped)███▊ | 3/5 [00:47<00:38, 19.13s/it]

zorropop commented 4 years ago

Where can I find the dumped core?

wangzheallen commented 4 years ago

Similar issue, the segment fault may come from https://github.com/vchoutas/torch-mesh-isect, however, not sure how to totally solve that with 'interpenetration' enabled. Already tried the ways 1.creating pure python virtual env instead of conda 2.compiling with sm_61 architecture specified

  1. make sure the cuda I am using is 10.1
  2. download cuda samples instead using local machine ones Finally it works with --interpenetration False, However, not sure about whether the performance is negatively affected or not.
zorropop commented 4 years ago

@wangzheallen Many thanks, that did the trick! Cudos!

wangzheallen commented 4 years ago

@vchoutas any hint on how to make the 'interpenetration' work?

wangzheallen commented 4 years ago

solved after upgrading system to Ubuntu18, which means ubuntu16 is nor working! should also refer to https://github.com/vchoutas/smplify-x/issues/3, as suggested by @huhai46312731

dimtzionas commented 4 years ago

Thanks for linking to the post of @huhai46312731 :)

As the read_me files say (here and here), this is the supported setup:

The code has been tested with Python 3.6, CUDA 10.0, CuDNN 7.3 and PyTorch 1.0 on Ubuntu 18.04.

wangzheallen commented 4 years ago

I think this issue can be closed.

wine3603 commented 4 years ago

I got this bug in both ubuntu 16 and 18, update to ubuntu 18 doesn't help. my python 3.7, cuda 10.0, cudnn 7.5 pytorch 1.1 and ubuntu 18