search

mks0601 / V2V-PoseNet_RELEASE

Official Torch7 implementation of "V2V-PoseNet: Voxel-to-Voxel Prediction Network for Accurate 3D Hand and Human Pose Estimation from a Single Depth Map", CVPR 2018
https://arxiv.org/abs/1711.07399
MIT License
377 stars 69 forks source link

Pre-computed centers for ITOP dataset seems problematic #34

Closed jb892 closed 5 years ago

jb892 commented 5 years ago

Hi Moon,

I came across a problem while inspecting the depth map and pre-computed center from ITOP dataset. Here is the result: dm1 dm2 dm3 dm4 dm5 The center is denoted by a red dot. However, it seems that the red dots are not the real center of the human body. Why is that? Did I miss anything?

mks0601 commented 5 years ago

Hi jb892,

I did not visualized the estimated centers on the ITOP dataset. If you used front-view centers to visualize above images and correctly transformed the centers to the pixel coordinates (using my code), there would be no problem. Note that the provided centers are estimation values, not GT values. Therefore, some errors might exist.

jb892 commented 5 years ago

Thanks for your reply!

I will take a closer look at your code. Cheers!

jb892 commented 5 years ago

Just found that the camera intrinsic parameters I used: fx, fy was wrong. I took them from the website of ITOP. The correct value should be fx = fy = 241.42 . correct_center

mks0601 commented 5 years ago

The camera parameters and pixel2world, world2pixel functions were from ITOP dataset website, and they worked well. The dataset has been changed? Can you tell me the address of the site? It seems https://www.albert.cm/projects/viewpoint_3d_pose/ is broken.

jb892 commented 5 years ago

I'm not sure now since I cannot remember where did I take those wrong figures. Maybe I mistakenly took them from the NYU hand dataset. BTW, the ITOP website is down now.

saileshprem commented 5 years ago

The new ITOP website is here: https://www.alberthaque.com/projects/viewpoint_3d_pose/