search

akashsengupta1997 / HuManiFlow

[CVPR 2023] Code repository for HuManiFlow: Ancestor-Conditioned Normalising Flows on SO(3) Manifolds for Human Pose and Shape Distribution Estimation
MIT License
74 stars 2 forks source link

Performance gap #5

Open noahcao opened 5 months ago

noahcao commented 5 months ago

Hi @akashsengupta1997 ,

I tried the evaluation code with your released default checkpoint but found a performance gap between what I got from the code and the performance reported in the paper.

This is the SSD-3D performance I got: Screenshot 2024-04-09 at 8 34 22 PM

This is the 3DPW performance I got: Screenshot 2024-04-09 at 8 34 28 PM

The results show a significant difference from what is shown in Table 3 -5 in the paper. Could you help to provide some clarification about it?