search

facebookresearch / InterHand2.6M

Official PyTorch implementation of "InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image", ECCV 2020
Other
676 stars 92 forks source link

questions about root depth #36

Open yanglilian opened 3 years ago

yanglilian commented 3 years ago

It seems that this project uses the absolute depth obtained from rootnet for all datasets in the evaluation procedure，so how do you train the rootnet for the hand project. As I know，rootnet is desinged for human pose. Can you release the training code for the rootnet in hand pose? thanks.

mks0601 commented 3 years ago

You don't have to change many things from RootNet repo. All you have to do are changing bbox_real to (300,300) and write data/InterHand26M/InterHand26M.py in the RootNet repo. (300,300) indicate hand scale of x- and y-axis in milimeter. You can write data/InterHand26M/InterHand26M.py in the RootNet repo by refering other dataloaders of the RootNet repo and this repo.

yanglilian commented 3 years ago

You don't have to change many things from RootNet repo. All you have to do are changing bbox_real to (300,300) and write data/InterHand26M/InterHand26M.py in the RootNet repo. (300,300) indicate hand scale of x- and y-axis in milimeter. You can write data/InterHand26M/InterHand26M.py in the RootNet repo by refering other dataloaders of the RootNet repo and this repo.

ok, I will try. Another question is have you ever tried merge the branch of absolute depth prediction into this project?

mks0601 commented 3 years ago

I found the code. Use this one. https://drive.google.com/drive/folders/1dYCgmHqW1qoD3osoB_Z4XWiaKcG7__-8?usp=sharing

You can set the path things by reading the README of this repo.

mks0601 commented 3 years ago

No I haven't. RootNet is my previous work and it can be applied to 3D body/hand pose/shape estimation.

yanglilian commented 3 years ago

I found the code. Use this one. https://drive.google.com/drive/folders/1dYCgmHqW1qoD3osoB_Z4XWiaKcG7__-8?usp=sharing

You can set the path things by reading the README of this repo.

ok, I will try. Thanks

shreyashampali commented 3 years ago

hi @mks0601 Thanks for this great work and dataset. I had a question regarding the MPJPE numbers (Table 4 in ECCV paper) in your paper for the interHands dataset. Are these numbers calculated using RootNet output or using ground truth root depths?

mks0601 commented 3 years ago

All numbers in the paper are obtained using RootNet.

ChiShao commented 3 years ago

I found the code. Use this one. https://drive.google.com/drive/folders/1dYCgmHqW1qoD3osoB_Z4XWiaKcG7__-8?usp=sharing

You can set the path things by reading the README of this repo.

hi @mks0601 I run train.py in InterHand2.6M-RootNet, but there is an error : [Errno 2] No such file or directory: '../data/RHD/data/RHD_training.json' Could you tell me how to get this file？