eth-siplab / AvatarPoser

Official Code for ECCV 2022 paper "AvatarPoser: Articulated Full-Body Pose Tracking from Sparse Motion Sensing"
https://siplab.org/projects/AvatarPoser
MIT License
277 stars 46 forks source link

Input points for testing and training #26

Closed Recialhot closed 5 months ago

Recialhot commented 6 months ago

Testing and training are based on 3 points. I would like to change it to 5 points (adding points with two feet). Where should I modify it?

jiaxi-jiang commented 5 months ago

Hey, you can simply change the data preprocessing code in prepare_data.py to add also foot pose, and increase the number of inputs of the network

Recialhot commented 5 months ago

嘿,你可以简单地在prepare_data.py中更改数据预处理代码,以添加脚姿势,并增加网络的输入数量

when i use z to up, wherever x and y are facing, it will cause this:

body_919

only when i use z to forward, x to left, y to up, it will cause this, but it not stand and Lifting feet is easy to fluctuate. when i see dataset_amass or your dataset_tracking, the data always z to up, but final result is right. my data z to up is not success. That's why? Thanks

body_100