search

Gengzigang / PCT

This is an official implementation of our CVPR 2023 paper "Human Pose as Compositional Tokens" (https://arxiv.org/pdf/2303.11638.pdf)
MIT License
315 stars 21 forks source link

add the other dataset to improve the performance #15

Open wxhqw opened 1 year ago

wxhqw commented 1 year ago

Have you add the CrowdPose, OCHuman, SyncOCC dataset to train the pct model ,try to improve the performance of 2d pose estimation

Gengzigang commented 1 year ago

Hello, thank you for your interest in our work. Due to the inconsistent annotation methods across these datasets, we have not attempted to train on multiple datasets jointly.