File size: 1,936 Bytes
e524557 | 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 | ---
license: mit
pretty_name: MapFly-13K
task_categories:
- robotics
tags:
- uav
- aerial-navigation
- visual-navigation
- benchmark
size_categories:
- 10K<n<100K
---
# MapFly-13K
13,300 validated UAV episodes over 15 Unreal Engine scenes for
**[MapFly](https://github.com/GuYan66/MapFly)**: prior-map-guided aerial visual
navigation. Each episode is a start, a goal, an expert route, first-person RGB,
and north-up OSM maps. Code, evaluation and MapFly-Agent live in the GitHub
repository; Linux UE packages and scene bundles are
[`EzGuYan/MapFly_DataGen`](https://huggingface.co/datasets/EzGuYan/MapFly_DataGen).
## Download
Archives are split zip sets under `original_data/<scene>.zip` plus `.z01`, `.z02`, …
The Hugging Face dataset viewer cannot open multi-disk zips; unpack them locally:
```bash
# one scene, e.g. smallcity
7z x original_data/smallcity.zip -o data/mapfly13k
# or: zip -s 0 original_data/smallcity.zip --out /tmp/smallcity-joined.zip && unzip /tmp/smallcity-joined.zip -d data/mapfly13k
```
Point MapFly's `configs/local.yaml` at that directory (`dataset_root`).
## Layout after unzip
```text
<data/mapfly13k>/
<scene>/
run.json
derived/bev.npz
<scene>_000000/{episode.json, obs/, maps/osm/, maps/osm_start_goal/, ...}
```
The Train / Test Seen / Test Unseen split is `seen12_v1`: 12 seen scenes train on
a sorted prefix (9,660 episodes) and hold out the rest (2,140); industrialcity,
laketown and moderncity2 are Test Unseen in full (1,500). The same file is
`splits/seen12_v1.json` here and `configs/splits/seen12_v1.json` in the GitHub
repo.
Only OSM maps are in this dataset. Satellite and markers-only maps are rendered
from the scene bundles (`MapFly_DataGen`, `bundles/`).
## License
MIT for this dataset. The UE scenes used to capture it keep their own Unreal
Engine / Marketplace licenses; the packages on `MapFly_DataGen` are for research
use with this benchmark.
|