About the performance of SMPL values of Human3.6M dataset

mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020

MIT License

708 stars 130 forks source link

About the performance of SMPL values of Human3.6M dataset #102

Open Ironbrotherstyle opened 2 years ago

Ironbrotherstyle commented 2 years ago

Thank you for sharing your great work. I have been working on making my own dataset with 3D joints and SMPL parameters. The data provided by you link is impressive. However, I am not sure if the provided data used 3D joints in training as your new paper are based on that. Did the data provided by you link1 link2 use 3D joints? The question is coming from the phenomenon that my try on Smplify-X with 2D supervision dose not show performance as yours.

mks0601 commented 2 years ago

Yes I ran SMPLify-X on GT 3D joints.

Ironbrotherstyle commented 2 years ago

Thank you.