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

Dataset issue #53

Closed Fazlik995 closed 4 years ago

Fazlik995 commented 4 years ago

Hi, Gyeongsik

Thanks for sharing your awesome work.

In your paper you used depth images with 3D joint locations.

I wonder can i apply your method for depth images without 3D joint locations(only depth image)?

Thanks

mks0601 commented 4 years ago

Do you mean the V2V-PoseNet can be trained in a self-supervised way? For 3D hand shape estimation, I think so. However, for 3D hand pose estimation, I'm not sure. You may want to see this paper.

Fazlik995 commented 4 years ago

Thank you.

I have one more question

Currently, I am working on gesture recognition and my datasets are DHG-14/28 and Shrec'17.

They both contain depth images(whole body, not only hand). However, dataset include 3D coordinates of hand.

I wanna re-extract 3D coordinates of hand using your method. But you implemented your method for datasets which contain only hands and when you voxelize, u only voxelized hand itself.

I am not sure dataset with whole body(not only hand) is suitable for your method, bce ig i voxelize i voxelize whole body not only hand.

What do you think, is it possible to implement your method for my datasets.

Sorry if I could not explain properly my issue.

I just wanna know ur opinion.

mks0601 commented 4 years ago

I think you should get some bounding boxes of hand and fed the region to my model?