I have a question in your hand detection module

[mks0601]

Hi thanks for sharing your code

I have a question about your hand detection module which is used in most of the hand pose estimation papers recently.

According to 96 line of training/GetH5DataNYU.py,

depth = CropImage(depth, joint_uvd[id, 34])

you used joint_uvd[id, 34] for com of CropImage function.

So, I`m curious about whether you used groundtruth palm position(joint_uvd[id, 34]) when you crop the hand from original depth image even in test stage or not.

[xingyizhou]

Hi mks0601,
Sorry for the late reply. Yes, we use the groundtruth palm position to crop the image when testing.

[mks0601]

Is it possible for you to say 'used same pre-processing as Deep prior paper' as in your paper?
Because they did not use GT position as center of mass as you can see in 603~616 line of https://github.com/moberweger/deep-prior/blob/master/src/data/importers.py

[xingyizhou]

Hi mks0601,
Thanks very much for pointing out the problem. It should be my mistake not to notice this when releasing this training code. Actually we use exact DeepPrior's code to produce their pickle file and then convert it to hdf5 file for the experiments in our paper. So you should be still able to reproduce our results.

Best,
Xingyi

[popper0912]

Is it possible for you to say 'used same pre-processing as Deep prior paper' as in your paper?
Because they did not use GT position as center of mass as you can see in 603~616 line of https://github.com/moberweger/deep-prior/blob/master/src/data/importers.py

他们也用了 就在你 603 - 616， 你测试了吗 这个点如果在测试的时候与训练时候不一致 结果对不对？

[mks0601]

I don't know how to read Chiness.