# IMHD$`^2`$: Inertial and Multi-view Highly Dynamic human-object interactions Dataset
[](https://arxiv.org/abs/2312.08869) [](https://afterjourney00.github.io/IM-HOI.github.io/)
> **I'M HOI: Inertia-aware Monocular Capture of 3D Human-Object Interactions**
> *Chengfeng Zhao, Juze Zhang, Jiashen Du, Ziwei Shan, Junye Wang, Jingyi Yu, Jingya Wang, Lan Xu\**
****
## Dataset Features
IMHD$`^2`$ is featured by:
- Human motion annotation in SMPL-H format, built on [EasyMocap](https://github.com/zju3dv/EasyMocap/tree/master)
- Object motion annotation, built on [PHOSA](https://github.com/facebookresearch/phosa)
- Well-scanned object geometry, using [Polycam](https://poly.cam/)
- Object-mounted IMU sensor measurement, using [Movella DOT](https://www.movella.com/products/wearables/movella-dot)
- 32-view RGB videos & instance-level segmentations, built on [SAM](https://github.com/facebookresearch/segment-anything), [Track-Anything](https://github.com/gaomingqi/Track-Anything) and [XMem](https://github.com/hkchengrex/XMem)
- 32-view 2D&3D human keypoints detection, using [ViTPose](https://github.com/ViTAE-Transformer/ViTPose) and [MediaPipe](https://github.com/google/mediapipe)
## Dataset Structure
```
data/
|--calibrations/ # camera intrinsics and world-to-cam extrinsics
|--object_templates/ # raw and downsampled geometry
|--imu_preprocessed/ # pre-processed IMU signal
|--keypoints2d/ # body keypoints in OP25 format and hand keypoints in MediaPipe format
|--keypoints3d/ # body keypoints in OP25 format and hand keypoints in MediaPipe format
|--video_release/ # raw videos from 32 multiple views
|--mask_release/ # human and object separate segmentations from 32 multiple views
|--ground_truth/ # human motion in SMPL-H format and rigid object motion
|----/
|------/
|--------/
|----------gt___.pkl
```
All sub-folders have the similar detailed structure as the shown one of ground truth. Particularly, since motion annotations of some part in some sequence are not ideal, there may exist several `.pkl` files under one sequence folder. To parse the file name meaning of leaf `.pkl` files, here is an example: `gt_0_10_100.pkl: the first motion part which starts from frame_10 and ends at frame_100`.
### How to use
Please refer to our [GitHub](https://github.com/AfterJourney00/IMHD-Dataset).
## Citation
If you find our data or paper helps, please consider citing:
```bibtex
@inproceedings{zhao2024imhoi,
title={I'm hoi: Inertia-aware monocular capture of 3d human-object interactions},
author={Zhao, Chengfeng and Zhang, Juze and Du, Jiashen and Shan, Ziwei and Wang, Junye and Yu, Jingyi and Wang, Jingya and Xu, Lan},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={729--741},
year={2024}
}
```
## Acknowledgement
This work was supported by National Key R\&D Program of China (2022YFF0902301), Shanghai Local college capacity building program (22010502800). We also acknowledge support from Shanghai Frontiers Science Center of Human-centered Artificial Intelligence (ShangHAI).
We thank Jingyan Zhang and Hongdi Yang for settting up the capture system. We thank Jingyan Zhang, Zining Song, Jierui Xu, Weizhi Wang, Gubin Hu, Yelin Wang, Zhiming Yu, Xuanchen Liang, af and zr for data collection. We thank Xiao Yu, Yuntong Liu and Xiaofan Gu for data checking and annotations.
## Licenses
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.