search

hongsukchoi / TCMR_RELEASE

Official Pytorch implementation of "Beyond Static Features for Temporally Consistent 3D Human Pose and Shape from a Video", CVPR 2021
MIT License
274 stars 39 forks source link

14 joints defined by Human3.6M are used for computing PA-MPJPE and MPJPE #39

Open Oruka94 opened 1 year ago

Oruka94 commented 1 year ago

Hello. You mentioned in your paper that you used 14 joints defined by Human3.6M for computing PA-MPJPE and MPJPE. I wonder those joints are which of the 17 joints in Human3.6M? Did you exclude H36M_NAMES[15] = 'Head', H36M_NAMES[14] = 'Neck/Nose', and H36M_NAMES[12] = 'Spine'?

yangjihao commented 1 year ago

I also want to know which 14 joints were used in evaluation.

hongsukchoi commented 1 year ago

I followed the previous literature

Please refer to here: https://github.com/hongsukchoi/Pose2Mesh_RELEASE/blob/d2c031d6ec7ab409f5981d7bef0b23444ec40e2e/data/Human36M/dataset.py#L51

Oratacth commented 1 year ago

I followed the previous literature

Please refer to here: https://github.com/hongsukchoi/Pose2Mesh_RELEASE/blob/d2c031d6ec7ab409f5981d7bef0b23444ec40e2e/data/Human36M/dataset.py#L51

When I validate, why does this happen

Traceback (most recent call last): File "train.py", line 138, in main(cfg) File "train.py", line 130, in main debug_freq=cfg.DEBUG_FREQ, File "D:\PROJECT\TCMR\lib\core\trainer.py", line 338, in fit self.validate() File "D:\PROJECT\TCMR\lib\core\trainer.py", line 300, in validate target_j3d = target['kp_3d'].view(-1, n_kp, 3).cpu().numpy() RuntimeError: shape '[-1, 49, 3]' is invalid for input of size 1344

hongsukchoi commented 1 year ago

It seems you have downloaded wrong data