search

mks0601 / NeuralAnnot_RELEASE

3D Pseudo-GTs of "NeuralAnnot: Neural Annotator for 3D Human Mesh Training Sets", CVPRW 2022 Oral.
173 stars 11 forks source link

Question about results replacing only 3d annotations used in SPIN #8

Closed MooreManor closed 2 years ago

MooreManor commented 2 years ago

@mks0601 Hello! NeuralAnnot is a wonderful work. I have a question about the experiment.

For SPIN used 3D dataset labels, that is, h36m mosh labels and mpi-inf-3dhp_mview_fits, will substituting NeuralAnnot pseudo labels with them raise the score (i.e. indirect 3d annotation error) while keeping ITW datasets labels with the same 2D GT annots without GT SMPL supervision?

mks0601 commented 2 years ago

I checked that their h36m and mpi-inf-3dhp fits are worse than NeuralAnnot's fits.

MooreManor commented 2 years ago

@mks0601 Is the comparision conducted through indirect 3d annotation error or visualization?

mks0601 commented 2 years ago

I checked the direct 3D annotation error

MooreManor commented 2 years ago

@mks0601 Is it possible to give the detailed number of the direct 3D annotation error between the neuralAnnot and the original one? Thanks!

mks0601 commented 2 years ago

Sorry I don't remember exact numbers.. But I remember there was some meaningful gap

MooreManor commented 2 years ago

@mks0601 Then how about the indirect 3d annotation error if you have done related experiments? Besides, do you try to compare the quality of annots with the latest CLIFF on ITW dataset?

mks0601 commented 2 years ago

For both, I haven't done related experiments. Sorry about this

MooreManor commented 2 years ago

@mks0601 Thanks for your patient and in-time reply! I am clear now :) 👍