search

bruceyo / GLA-GCN

2D to 3D human pose lifting
66 stars 2 forks source link

great at gt but worse at cpn #6

Open Aerikervalid opened 11 months ago

Aerikervalid commented 11 months ago

Hi， As in your case，My model is great at gt but worse at cpn，Adding some module can make gt better but cpn gets worse.What does the author think about this situation? ls there any way to solve it. Looking forward to your reply! Thanks！

bruceyo commented 11 months ago

Hi， As in your case，My model is great at gt but worse at cpn，Adding some module can make gt better but cpn gets worse.What does the author think about this situation? ls there any way to solve it. Looking forward to your reply! Thanks！

Hi, thank you for the interest! Recent methods usually integrate 2D pose detection together with 3D pose reconstruction to achieve good performance. Based on our experiments, I do not think achieving good results on CPN is a proper problem setup. This issue has been discussed in the paper.