File size: 3,190 Bytes
e8ad724
c036393
39334eb
d09e1ba
 
39334eb
d09e1ba
 
 
 
39334eb
d09e1ba
 
 
 
 
e8ad724
c036393
 
 
d09e1ba
 
 
 
 
c036393
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d09e1ba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
90
91
92
93
94
95
96
97
---
license: mit
size_categories:
- n<1K
pretty_name: AniGen Sample Data
tags:
- 3d
- image
task_categories:
- image-to-3d
configs:
- config_name: default
  default: true
  data_files:
  - split: train
    path: samples.csv
---

# AniGen Sample Data

[Paper](https://huggingface.co/papers/2604.08746) | [Project Page](https://yihua7.github.io/AniGen_web/) | [GitHub](https://github.com/VAST-AI-Research/AniGen)

This directory is a compact example subset of the AniGen training dataset, as presented in the paper [AniGen: Unified $S^3$ Fields for Animatable 3D Asset Generation](https://huggingface.co/papers/2604.08746). 

AniGen is a unified framework that directly generates animate-ready 3D assets conditioned on a single image by representing shape, skeleton, and skinning as mutually consistent $S^3$ Fields.

## What Is Included

- 10 examples
- 10 unique raw assets
- Full cross-modal files for each example
- A subset `metadata.csv` with 10 rows

The retained directory layout follows the core structure of the reference test set:

```text
raw/
renders/
renders_cond/
skeleton/
voxels/
features/
metadata.csv
statistics.txt

latents/ (encoded by the trained slat auto-encoder)
ss_latents/  (encoded by the trained ss auto-encoder)
```

## How To Read One Example

Each row in `metadata.csv` corresponds to one example identifier in the `sha256` column. In practice this value is the sample key used across modalities.

For a row with sample key `<file_identifier>`:

- raw asset: `local_path` field, for example `raw/<raw_file>`
- rendered views: `renders/<file_identifier>/`
- conditional rendered views: `renders_cond/<file_identifier>/`
- skeleton files: `skeleton/<file_identifier>/`
- voxel files: `voxels/<file_identifier>.ply` and `voxels/<file_identifier>_skeleton.ply`
- image feature: `features/dinov2_vitl14_reg/<file_identifier>.npz`
- mesh latents: files under `latents/*/<file_identifier>.npz`
- structure latents: files under `ss_latents/*/<file_identifier>.npz`

## Sample Usage (Training)

According to the [official repository](https://github.com/VAST-AI-Research/AniGen), you can use this data for training by following these stages:

```bash
# Stage 1: Skin AutoEncoder
python train.py --config configs/anigen_skin_ae.json --output_dir outputs/anigen_skin_ae

# Stage 2: Sparse Structure DAE
python train.py --config configs/ss_dae.json --output_dir outputs/ss_dae

# Stage 3: Structured Latent DAE
python train.py --config configs/slat_dae.json --output_dir outputs/slat_dae

# Stage 4: SS Flow Matching (image-conditioned generation)
python train.py --config configs/ss_flow_duet.json --output_dir outputs/ss_flow_duet

# Stage 5: SLAT Flow Matching (image-conditioned generation)
python train.py --config configs/slat_flow_auto.json --output_dir outputs/slat_flow_auto
```

## Citation

```bibtex
@article{huang2026anigen,
  title     = {AniGen: Unified $S^3$ Fields for Animatable 3D Asset Generation},
  author    = {Huang, Yi-Hua and Zhou, Zi-Xin and He, Yuting and Chang, Chirui
               and Pu, Cheng-Feng and Yang, Ziyi and Guo, Yuan-Chen
               and Cao, Yan-Pei and Qi, Xiaojuan},
  journal   = {ACM SIGGRAPH},
  year      = {2026}
}
```