https://github.com/bryanbocao/tic

Official Pytorch Implementation of Transformer IMU Calibrator

https://github.com/bryanbocao/tic

Repository

Official Pytorch Implementation of Transformer IMU Calibrator

https://github.com/bryanbocao/TIC/blob/main/

Transformer IMU Calibrator: Dynamic On-body IMU Calibration for Inertial Motion Capture


    Chengxu Zuo1 
    Jiawei Huang1 
    Xiao Jiang1 
    Yuan Yao1 
    Xiangren Shi4 
    Rui Cao1 
    Xinyu Yi2 
    Feng Xu2 
    Shihui Guo1* 
    Yipeng Qin3 


    1Xiamen University   2Tsinghua University   3Cardiff University   4Bournemouth University



    *Corresponding Author



     SIGGRAPH 2025 Best Papers




[//]: # ([Paper] )

[Paper]
[Fast-Forward Video]
[Demo]
  




![](figs/teaser.jpg)
Implementation of our SIGGRAPH 2025 paper "Transformer IMU Calibrator: Dynamic On-body IMU Calibration for Inertial Motion Capture". Including network weights, training and evaluation scripts.

[train.py](./train.py): TIC Network training.

[eval.py](./eval.py): Run our dynamic calibration on dataset and calculate OME, AME and R_G'G/R_BS Error.

## Synthesized Dataset for training

Prepare AMASS and DIP-IMU dataset then run [data_preprocess.py](./data_preprocess.py).

[//]: # (1. Download required training data at xxxx.)

[//]: # (2. Copy all data in folder: [root/data_train])

[//]: # ()
[//]: # (*Note: We expand synthesized head IMU acc data with 14 different vertices on head mesh (head_acc.pt), thus covering acc variances on different IMU location when rotating head.)


## TIC Dataset
The TIC dataset is available at https://www.dropbox.com/scl/fo/ggrvm8x2xjhu1m0pjomc9/ADClW3gbt4swggoulhndBKA?rlkey=bagguhrnze7fdvgr2toggce0v&st=p3fj8g1e&dl=0.

The data was collected from 5 subjects (s1~s5).
For each subject, the dataset provides:
1. **acc.pt**----Acceleration of 6 on-body IMU, calibrated by static calibration at begin.
2. **rot.pt**----Orientation of 6 on-body IMU, calibrated by static calibration at begin.
3. **pose.pt**----SMPL pose captured by NOKOV System (use optical tracker).
4. **trans.pt**----Global body translation (location).
5. **drift.pt**----Absolute coordinate drift of 6 on-body IMU.
6. **offset.pt**----Measurement offset of 6 on-body IMU.
7. **acc_gt.pt**----GT IMU acceleration captured by NOKOV System (use optical tracker).

*Note 1: IMU order: left forearm, right forearm, left lower leg, right lower leg, head, hip

*Note 2: All data are in SMPL frame. 

## Acknowledgement

Some of our codes are adapted from [PIP](https://github.com/Xinyu-Yi/PIP).
The SMPL_MALE model is download from https://smpl.is.tue.mpg.de/.

[//]: # (## Citation)

[//]: # ()
[//]: # (If you find this project helpful, please consider citing us:)

Owner

GitHub Events