This repository contains the implementation of our paper, Garment Inertial Denoiser: Endowing Accurate Motion Capture via Loose IMU Denoiser, including pretrained weights, training scripts, and evaluation code.
- train_per_imu.py: trains Location-Specific Denoiser expert modules, each specialized for one IMU placement.
- train_fuse.py: trains the Cross-wear Fusion model, initialized from the retained Location-Specific Denoiser checkpoints.
- denoise.py: generates denoised IMU sequences with the trained model.
- eval.py: evaluates saved denoised IMU data with MAE.
The GID dataset is available at:
Google Drive: https://drive.google.com/drive/folders/1tp6yjy3AbLiOAsud96pyHmdJqPSqFo6R?usp=sharing
Baidu Cloud: https://pan.baidu.com/s/1LDw5Z8BhQLjhqbE4RepKvg?pwd=7hap