---
license: cc-by-nc-4.0
language:
- en
tags:
- satellite-imagery
- remote-sensing
- hierarchical
- benchmark
- pyramid-generation
- dense500
size_categories:
- 10K<n<100K
pretty_name: Dense500
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
dataset_info:
  features:
  - name: image
    dtype: image
  - name: location
    dtype: string
  - name: site_id
    dtype: string
  - name: tag
    dtype: string
  - name: continent
    dtype: string
  - name: site_lon
    dtype: float64
  - name: site_lat
    dtype: float64
  - name: zoom
    dtype: int32
  - name: tile_x
    dtype: int32
  - name: tile_y
    dtype: int32
  - name: site_start_zoom
    dtype: int32
  - name: site_leaf_zoom
    dtype: int32
  - name: site_depth
    dtype: int32
  - name: is_root
    dtype: bool
  - name: is_leaf
    dtype: bool
  - name: zoom_window
    dtype: int32
  - name: site_index_in_window
    dtype: int32
  splits:
  - name: train
    num_bytes: 1087489040
    num_examples: 42500
  download_size: 1087489040
  dataset_size: 1087489040
---

# Dense500

Fully observed satellite tile pyramids for multi-scale tile completion. Each
site is a complete depth-4 quadtree: 85 tiles of 256×256 at four consecutive
zoom levels (1 + 4 + 16 + 64), so any seed configuration can be completed and
scored against ground truth at every level. This is the benchmark of Genesis
(ACM SIGSPATIAL 2026). A companion retrieval split with object annotations and
captions is in
[`MVRL/dense500-retrieval`](https://huggingface.co/datasets/MVRL/dense500-retrieval)
and
[`MVRL/dense500-retrieval-instances`](https://huggingface.co/datasets/MVRL/dense500-retrieval-instances).

## Construction

Site centers were sampled worldwide with a population prior and continent
stratification, restricted to populated areas. Sites are organized in five
zoom windows, each sliding a four-level span up the pyramid; for each site,
every tile between its root zoom and its leaf zoom was collected. Sites whose
imagery source changes between adjacent zoom levels were removed, so every
site is visually coherent across its four zooms.

| Window | Root zoom | Leaf zoom |
| --- | --- | --- |
| W1 | 11 | 14 |
| W2 | 12 | 15 |
| W3 | 13 | 16 |
| W4 | 14 | 17 |
| W5 | 15 | 18 |

## Schema

Each row is one tile.

| Column | Type | Description |
| --- | --- | --- |
| `image` | image | 256×256 JPEG |
| `location` | string | Tile key `"{zoom}_{x}_{y}"` (Web Mercator) |
| `site_id` | string | Groups all 85 tiles of one quadtree |
| `tag` | string | `naeu_urban` or `global_urban` |
| `continent` | string | NA, EU, Asia, Africa, SA, Oceania |
| `site_lon`, `site_lat` | float64 | Site center (degrees) |
| `zoom`, `tile_x`, `tile_y` | int32 | Web Mercator tile coordinates |
| `site_start_zoom`, `site_leaf_zoom` | int32 | Root and leaf zoom of the site |
| `site_depth` | int32 | Always 4 |
| `is_root`, `is_leaf` | bool | Tile position in its site |
| `zoom_window` | int32 | 1 to 5, W1 to W5 |
| `site_index_in_window` | int32 | Site index within its window |

## Loading

```python
from datasets import load_dataset

ds = load_dataset("MVRL/dense500", split="train")

from collections import defaultdict
by_site = defaultdict(list)
for row in ds:
    by_site[row["site_id"]].append(row)

w3 = ds.filter(lambda r: r["zoom_window"] == 3)
```

## Citation

```bibtex
@inproceedings{khanal2026genesis,
  author    = {Khanal, Subash and Cui, Yangzhi and Cher, Daniel and Xing, Eric and Wei, Brian and Sastry, Srikumar and Jacobs, Nathan},
  title     = {Genesis: A Generative Engine for Hierarchical Satellite Image Synthesis},
  booktitle = {ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (ACM SIGSPATIAL)},
  year      = {2026},
  doi       = {10.1145/3841645.3843313}
}
```