Nicholasli1995 / EvoSkeleton

Official project website for the CVPR 2020 paper (Oral Presentation) "Cascaded deep monocular 3D human pose estimation wth evolutionary training data"
https://arxiv.org/abs/2006.07778
MIT License
333 stars 44 forks source link

Source of 2D keypoints when eval on 3dhp dataset #78

Open ZhangRenkai opened 1 year ago

ZhangRenkai commented 1 year ago

May I ask a question? When doing this cross dataset experiment, is the 2D GT coordinates in the 3dhp dataset as the input or the coordinates estimated by the 2D pose estimator hrnet as the input. image

I just use 2D GT annots of 3dhp dataset as the 2d input, and the results are as below. (The 2Dto3D model is trained on this config: stage 2 block 3 evolve: generation_6) image And then I use 2D HRNet outputs as the 2d input, and the results are as below. (The 2d model is final_state.pth you mentioned here. https://github.com/Nicholasli1995/EvoSkeleton/blob/master/docs/2DHPE.md) image

Nicholasli1995 commented 1 year ago

Hi, as explained in https://github.com/Nicholasli1995/EvoSkeleton/issues/8, the 2D coordinates annots of the 3dhp dataset are used.