search

jfzhang95 / PoseAug

[CVPR 2021] PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation, (Oral, Best Paper Award Finalist)
MIT License
368 stars 57 forks source link

the data preprocess pipeline #12

Closed gravitychen closed 3 years ago

gravitychen commented 3 years ago

Nice work! I draw the data preprocess pipeline here, can you help me to check its correctness

huamn3.6(17 world 3D joints) --> delete nose joint --> huamn3.6(16 world 3D joints) -->normalize 3D (/ 1000) -->huamn3.6( noramlized 16 world 3D joints) --> normalize camera parameter and perspective projection --> get normalized 2D pose

MPI: delete nose --> (16 2D pose) --> normalize 2D using camera resolution -->input to poseaug -->predicted MPI 3D <==MPJPE protocal 1, PCK, AUC=> GT MPI 3D, then the MPJPE protocal 1, PCK, AUC are recorded in your paper, Table2

Is this right? thank you very much~

Garfield-kh commented 3 years ago

Thanks for your interest. Here the MPI 2D/3D pose are 24 joints setting, we convert it to 16 joints setting by prepare_data_3dhp.py, with reason issue #4 The others are right. :)

gravitychen commented 3 years ago

OK, Thank you very much!,I got it