search

namepllet / HandOccNet

Offical pytorch implementation of "HandOccNet: Occlusion-Robust 3D Hand Mesh Estimation Network", CVPR 2022.
156 stars 30 forks source link

about the projection #11

Closed lzfff12 closed 2 years ago

lzfff12 commented 2 years ago

How to project the predicted 3D joints back into the image after cropping the image and then match the 2D joints?

namepllet commented 2 years ago

Our model predicts 3D joints in root-relative coordinate system.

So if you'd like to project it you should translate predicted 3D joints using GT root coordinate.

Then you can project the translated 3D joints using cam2pixel in here

(If you'd like to project into cropped image, you can apply affine transform to cropped image space)

anilesec commented 2 years ago

Hi @namepllet

I have an extended question:

by "GT root coordinate" do you mean ground truth 3d joints root coordinates of that image?

if this is the case, then there is no way to project estimated 3d joints onto image if we don't have GT 3d joints for an input image. Is my understanding correct?

Thanks!