search

mks0601 / 3DMPPE_ROOTNET_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
477 stars 65 forks source link

PoseNet #5

Closed surelyee closed 5 years ago

surelyee commented 5 years ago

hello, thanks for your good work. Can you make me know where is your posenet code and MPJPE in your paper iccv2019? I am looking for your reply. thanks.

mks0601 commented 5 years ago

https://github.com/mks0601/3DMPPE_POSENET_RELEASE

You can run the code with provided pre-trained model to get MPJPE.

surelyee commented 5 years ago

@mks0601 thanks, i run the code of test.py and only get MRPE, why no MPJPE? python test.py --gpu 0-1 --test_epoch 7 ; for h36m dataset

mks0601 commented 5 years ago

I think you ran RootNet

surelyee commented 5 years ago

@mks0601 I am sorry, i can't find the .py file to test RootNet, and the RootNet is used in model.py, but it's result only is coord. the coord meaning maybe is the root of a person , the coord is (x,y,z) and the shape is (1,3), is not the 17*3 jonit of one person. thanks.

mks0601 commented 5 years ago

Could you make it clear? Do you want RootNet or PoseNet? The RootNet localizes human root in camera-centered coordinate system. The output is (x,y,z) of human root joint, which means shape is (1,3). The PoseNet localizes human joints in root-relative coordinate system. The output is (x,y,z) of human joints, which means shape is (J,3).

mks0601 commented 5 years ago

For the RootNet, use this repo. For the PoseNet, use above link.

surelyee commented 5 years ago

@mks0601 thank you very much!! this is my wrong for the name.........thanks for your patient answer. thanks a lot.