search

mks0601 / 3DMPPE_POSENET_RELEASE

Official PyTorch implementation of "Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Image", ICCV 2019
MIT License
807 stars 147 forks source link

Questions about the evaluation on MuPoTS #106

Closed Juzezhang closed 2 years ago

Juzezhang commented 2 years ago

Hello, thank you for your amazing repo. I noticed that you back-project (x,y) pixel + z milimeter to (x,y,z) milimeter in test stage. In 3DPCKrel evaluation case, I am confused whether you use GT z or rootnet output z in your paper. It will be very useful if you could provide more details. Thank a lot.

mks0601 commented 2 years ago

Hi, I use RootNet output z.

Juzezhang commented 2 years ago

Thanks for reply! But I found that in your release code and pretrained model that 81.8 PCKrel maybe obtain by using the GT z which refer from the following code? Line 132 https://github.com/mks0601/3DMPPE_POSENET_RELEASE/blob/master/data/MuPoTS/MuPoTS.py#L132

I don't know whether I understand it incorrectly.

mks0601 commented 2 years ago

Here, root_cam is from rootnet output: https://github.com/mks0601/3DMPPE_POSENET_RELEASE/blob/3f92ebaef214a0eb1574b7265e836456fbf3508a/data/MuPoTS/MuPoTS.py#L91

Juzezhang commented 2 years ago

Here, root_cam is from rootnet output:

https://github.com/mks0601/3DMPPE_POSENET_RELEASE/blob/3f92ebaef214a0eb1574b7265e836456fbf3508a/data/MuPoTS/MuPoTS.py#L91

Here, root_cam is from rootnet output:

https://github.com/mks0601/3DMPPE_POSENET_RELEASE/blob/3f92ebaef214a0eb1574b7265e836456fbf3508a/data/MuPoTS/MuPoTS.py#L91

I got it, Thanks very much!