search

vye16 / slahmr

MIT License
441 stars 50 forks source link

about scipy version #21

Closed TnoobT closed 1 year ago

TnoobT commented 1 year ago

hi, bro. very nice work！I want to know what version of scipy you have installed. I installed scipy=1.11.0 and reported an error image

geopavlakos commented 1 year ago

We provide this file with the full details regarding the versions of the dependencies we use for our environment. The version for scipy is 1.9.1.

TnoobT commented 1 year ago

Thank you very much for your reply. I have another question. I tested my own video and when processing frame 7, the code got stuck. Then I debugged and found that the value in the graph was nan, which entered a dead loop. Do you have a way to handle this situation

image

geopavlakos commented 1 year ago

Could you share the video?

TnoobT commented 1 year ago

I also tested 022691 Mpii Test will also fall into a dead cycle. Is it something wrong with my environment image image image image

geopavlakos commented 1 year ago

Yes, that video should run without a problem, so it's probably the environment. You can take a look at the colab to identify differences compared to the default environment and the default installation procedure.

TnoobT commented 1 year ago

Thank you for your answer. I have another silly question. I saw in your paper that the SMPL parameters were predicted using humor networks. Have you done any network or loss optimization for the estimation of SMPL parameters

geopavlakos commented 1 year ago

Yes, we solve an optimization problem to estimate the SMPL parameters for each person. The HuMoR network is responsible for one of the objectives of this optimization, the motion prior part.

TnoobT commented 1 year ago

I ran the code using Colab, but this issue also occurred. Perhaps you need to modify Colab. image

I made the modifications and now it can work normally image

geopavlakos commented 1 year ago

Thanks for pointing that out! I have updated the colab environment.

TnoobT commented 1 year ago

hi, I have reinstalled the environment several times but still fell into a dead cycle. I would like to ask if the version of CUDA on your computer is 11.7, and my version is 11.1. I would like to upgrade my CUDA version

geopavlakos commented 1 year ago

We have indeed tested the demo using CUDA version 11.7, but I cannot be entirely sure if the CUDA version is responsible for the issue you are facing.

TnoobT commented 1 year ago

ok, thank you. i try it

TnoobT commented 1 year ago

Damn it, I know what the reason is. On my own computer, CUDA is 11.1, but in install.sh, CUDA 11.7 corresponds to Torch1.13. I installed Torch1.13, but the program did not report any errors! Then I installed the torch corresponding to cuda11.1 (the torch-scatter also needs to correspond, otherwise the subsequent DROID-SLAM will report an error), the code ran successfully, and finally solved the problem