hongsukchoi / Pose2Mesh_RELEASE

Official Pytorch implementation of "Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from a 2D Human Pose", ECCV 2020
MIT License
678 stars 69 forks source link

Estimation uncertainty problem #62

Open PiggyBiuBiuBiu opened 1 year ago

PiggyBiuBiuBiu commented 1 year ago

Hello, I have encountered some problems of inaccurate estimation. (I use coco joint_set)

  1. 2D joints and viewability(demo_pose2d.png) are symmetrical.(As shown in the following figure) image
  2. Mesh estimation is not accurate(As shown in the following figure) image image
  3. I also want to ask if your code can estimate 2d directly from the picture and then get the mesh?(I think maybe my 2D joints are inaccurate?) These are my estimated 2d joints(19*2). [ 6. 28.44444466] [ 4.66666651 29.33333397] [ 4.66666651 27.55555534] [ 5.33333349 32.88888931] [ 5.33333349 26.66666603] [12. 36.44444275] [13.33333302 25.77777863] [20.66666603 40. ] [22.66666603 24.88888931] [29.33333397 40. ] [30.66666603 23.11111069] [28.66666603 34.66666794] [28.66666603 28.44444466] [42. 34.66666794] [42.66666794 30.22222137] [56. 32.88888931] [55.33333206 31.11111069] [28.66666603 31.55555725] [12.66666603 31.11111069]

I think maybe the picture resolution is too low.(64*64),so I want to ask you if this is the reason, can you help me solve this problem?