File size: 2,996 Bytes
30bf06b
7013b28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30bf06b
7013b28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
---
license: mit
pretty_name: Humans with Collisions (HwC) Pose & Motion Dataset
task_categories:
- robotics
- other
tags:
- 3d-graphics
- smpl-h
- collision-resolution
- human-pose
- motion-analysis
configs:
- config_name: default
  data_files:
  - split: train
    path: data/dataset/train_list.csv
  - split: test
    path: data/dataset/test_list.csv
---

# Humans with Collisions (HwC) Pose & Motion Dataset

This dataset contains the training, evaluation, and benchmark data for the paper:  
**"PoseShield: Neural Collision Fields for Human Self-Collision Resolution (ECCV 2026)"**

- **Paper (arXiv):** [arXiv:2606.29686](https://arxiv.org/abs/2606.29686)
- **Code Repository:** [PoseShield on GitHub](https://github.com/Tencent-Hunyuan/HY-Motion-1.0) (or project repo)

---

## Dataset Structure

The repository contains two main groups of data structured under the `data/` directory:

### 1. HwC Pose Dataset (Single Poses)
Used for training the neural self-collision field and evaluating pose-level collision resolution.
* `data/dataset/train_list.csv` - List of training sample IDs.
* `data/dataset/test_list.csv` - List of testing sample IDs.
* `data/dataset/augmented_data/` - Folder containing self-colliding SMPL-H body poses (`.npz`) used as negative training inputs.
* `data/dataset/gt_data/` - Folder containing corresponding collision-free ground truth poses (`.npz`).
* `data/dataset_test/` - The HwC 500-pose benchmark subset used for single-pose collision resolution validation, containing body models (`.pkl`), mesh files (`.obj`), and visualization references (`.png`).

### 2. Motion Dataset (Motion Sequences)
Used for two-stage latent motion optimization and visual/numerical self-collision resolution benchmark.
* `data/motion_canonical/` - Folder containing the 100 canonical MotionFix self-intersecting human motion sequences (`.npy`).

---

## Usage Instructions

To use this dataset in your project, you can clone this repository directly or download the snapshot programmatically.

### Cloning via Git LFS

Make sure you have Git LFS installed to fetch the `.npz` and `.npy` files correctly:

```bash
git lfs install
git clone https://huggingface.co/datasets/ZYYY99/Humans_with_Collision
```

### Programmatic Download (Python)

You can download the entire folder structure programmatically using the `huggingface_hub` Python package:

```python
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="ZYYY99/Humans_with_Collision", 
    repo_type="dataset", 
    local_dir="data"
)
```

## Citation

If you use this dataset or the matching method in your research, please cite:

```bibtex
@article{li2026poseshield,
  title={PoseShield: Neural Collision Fields for Human Self-Collision Resolution},
  author={Li, Zhengyuan and Deng, Zeyun and Shen, Yifan and Gui, Liangyan and Xie, Miaolan and Campbell, Joseph and Gao, Xifeng and Wu, Kui and Pan, Zherong and Bera, Aniket},
  journal={arXiv preprint arXiv:2606.29686},
  year={2026}
}
```