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
724
stars
127
forks
source link
questions about quality of SMPLify-X fitting error 13.1mm #80
Hi, @mks0601,
Sorry to bother you. I've computed the fitting errors of your provided SMPLify-X fits, on the training set, the error is 25.72mm, and on the test set, the error is 31.24mm, instead of 13.1mm. Could you please help me figure out the reason?
gt pose3d is from joint_cam
fitted pose3d is computed by using the joint_regressor_h36mwith the mechcoordinates computed by the function get_smpl_coordusing shape, pose, and transin your code,
I've aligned both to their root.
It seemed that I didn't miss anything? So is it available for you to provide your evaluation code of which get 13.1mm?
Thank you in advance!
Hi, @mks0601, Sorry to bother you. I've computed the fitting errors of your provided SMPLify-X fits, on the training set, the error is 25.72mm, and on the test set, the error is 31.24mm, instead of 13.1mm. Could you please help me figure out the reason?
gt pose3d is from
joint_camfitted pose3d is computed by using the
joint_regressor_h36mwith themechcoordinates computed by the functionget_smpl_coordusingshape,pose, andtransin your code,I've aligned both to their root.
It seemed that I didn't miss anything? So is it available for you to provide your evaluation code of which get 13.1mm? Thank you in advance!
Best, Shirley
_Originally posted by @ShirleyMaxx in https://github.com/mks0601/I2L-MeshNet_RELEASE/issues/22#issuecomment-885412374_