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

about the MANO parameters #29

Open liwenssss opened 3 years ago

liwenssss commented 3 years ago

Hi , I noticed that your update MANO parameters is fitted to the world coordinate.But I think fitting the camera coordinate as the most hand dataset do is better. Can you tell me the benefit to fit the world coordinate ?

mks0601 commented 3 years ago

Basically, you can obtain camera-centered coordinates by applying the camera extrinsics. Fitting to world coordinates can reduce the file size by the number of the camera, because there needs only one MANO parameter for each time step. On the other hand, fitting to camera coordinates require MANO parameters for all views.

liwenssss commented 3 years ago

I have a question, a hand in different camera views has the same pose parameters except the first 3 global rotation parameters, am I right?

liwenssss commented 3 years ago

image and it seems something wrong.. I just re-download the annotation file and run the updated render demo..

mks0601 commented 3 years ago

sorry there was a bug (confusion between meter and milimeter). Now, it works!