search

yohanshin / WHAM

MIT License
645 stars 71 forks source link

How to reconstruct the 3D keypoints? #111

Open Nexisato opened 3 weeks ago

Nexisato commented 3 weeks ago

I have found that the below code can generate the correct 2d keypoints for image visulaization, so I think but joints3d + full_cam is the correct 3d keypoint represented in the camera frame. But the result is not, when I get very close to the camera the last axis of joints3d + full_cam is bigger than 1(which i think the unit is meter).

  full_joints2d = full_perspective_projection(
      joints3d,
      translation=full_cam,
      cam_intrinsics=cam_intrinsics,
  )
  output.full_joints2d = full_joints2d
  output.full_cam = full_cam.reshape(-1, 3)
  output.joints3d = joints3d

So how to reconstruct the 3D keypoints?

GuoPingPan commented 3 weeks ago

Such a great job. I also want to know how to get 3d keypoint in camera frame. Could you please give me some help? @yohanshin