Vassilis Choutas
Vassilis Choutas
Yes, all of them are in meters. The world coordinate comes from the scanner. The transformation to the camera frame is included in the EHF data, in the EHF_camera.txt file....
Can I close this issue?
This seems like a torchvision issue. What version are you using?
What is your environment? Have you tried setting your EGL device ID with the env var EGL_DEVICE_ID ?
The warnings can be safely ignored. Was anything else printed? What code did you use in Colab?
@penincillin The code in the repo only has the vertex-to-vertex error computation. For the point-to-surface metric, we first do a rigid alignment between the ground truth and estimated mesh and...
Which dataset? What command did you use?
@joshuajano We released code in the smplx repository to transfer SMPL to SMPL-X parameters. That's what we used to evaluate for evaluate to get SMPL-X V2V numbers. Can you share...
I will push a fix to the demo and inference scripts. For now you should just add the camera scale and translation in the output [in this function](https://github.com/vchoutas/expose/blob/dffc38d62ad3817481d15fe509a93c2bb606cb8b/demo.py#L170).
@chaytonmin Do you have some specific example? The results will change due to the different boxes. Sometime the RCNN might give you a partial body detection, e.g. a lower body,...