File size: 3,369 Bytes
23872fe 60733d9 23872fe | 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 98 99 100 101 102 103 104 | ---
license: mit
task_categories:
- robotics
- video-classification
tags:
- robotics
- world-model
- video-prediction
- imitation-learning
- depth-estimation
- affordance
- libero
size_categories:
- 1K<n<10K
---
# DeVA — LIBERO (processed)
Processed **LIBERO** data for [DeVA](https://github.com/Mq-Zhang1/deva), a robot video-world
model with affordance + depth physical guidance. Derived from
[Lifelong-Robot-Learning/LIBERO](https://github.com/Lifelong-Robot-Learning/LIBERO) —
**please cite the original benchmark as well**.
| | |
|---|---|
| Episodes | 2,000 |
| Frames | 338,575 |
| Video | 432 × 768, 16 fps (2 views tiled side by side) |
| Views | `agentview`, `eye_in_hand` |
| Action dim | 7 (absolute) |
| Proprio state | not included |
| Size | 9.6 GB |
## Download
```bash
hf download mengqz9/deva_libero --repo-type dataset --local-dir datasets/deva_libero
```
## Layout
```
deva_libero/
├── dataset_info.json # view names, tile grid, affordance/depth encodings
├── videos/ # <episode>.mp4 tiled multi-view RGB
├── metas/ # <episode>.txt language instruction
├── t5_xxl/ # <episode>.pickle precomputed T5-XXL embedding [n_tok, 1024]
├── action.json # {episode: [[a_0..a_6], ...]} len == video frames
├── data_mapping.json # episode -> source task
├── norm_stats.json # action normalization statistics
├── affordance/ # <episode>.npz physical guidance (optional at train time)
└── depth/ # <episode>.npz physical guidance (optional at train time)
```
`videos/` tiles the views on the grid given by `dataset_info.json::tile`:
```
[[agentview, eye_in_hand]]
```
## Affordance / depth `.npz`
Both are keyed by **canonical view name** (never by tile position), one array per view:
| | affordance | depth |
|---|---|---|
| key | `<view>` | `<view>` |
| dtype | `float16` | `float16` |
| shape | `[T, 128, 128]` | `[T, 128, 128]` |
| range | raw, un-normalized | `[0, 1]` |
| note | loader resizes to 128×128, then per-frame max-normalizes | per-view per-clip min-max applied at export |
`T` equals the episode's video frame count, so aux arrays index by absolute frame.
The validity mask is derived from the tile grid (`null` cells are invalid) and is not stored.
**Depth source** — DepthAnything (ViT-L) disparity, converted as `depth = 1 / disparity`;
**higher value = farther**. The far tail is clipped at the 98th percentile before
per-view min-max, so ~2% of pixels saturate at 1.0.
**Affordance source** — simulator oracle contacts rendered as Gaussian heatmaps.
## Citation
If you use this data, please cite DeVA and LIBERO:
```bibtex
@article{zhang2026deva,
title = {{DeVA}: Decoupled Video-Action Model with physical guidance for robot policy learning},
author = {Zhang, Mengqi and Khose, Sahil and Kareer, Simar and
Song, Yuchen and Jain, Unnat and Hoffman, Judy},
journal = {arXiv preprint arXiv:2607.24159},
year = {2026}
}
@article{liu2023libero,
title = {LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},
author = {Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and
Liu, Qiang and Zhu, Yuke and Stone, Peter},
journal = {NeurIPS Datasets and Benchmarks},
year = {2023}
}
```
|