Dataset Viewer
Auto-converted to Parquet Duplicate
image
imagewidth (px)
512
512
id
stringlengths
4
4
bbox_rd_new
listlengths
4
4
crs
stringclasses
1 value
gsd_m
float64
1
1
stuff_c34_mask
stringlengths
28
28
n_stuff_c34
int64
1
3
things_bag_mask
stringlengths
29
29
n_things_bag
int64
1
55
t000
[ 175088, 401432, 175600, 401944 ]
EPSG:28992
1
masks/dev/t000_stuff_c34.png
1
masks/dev/t000_things_bag.png
19
t001
[ 229064, 513408, 229576, 513920 ]
EPSG:28992
1
masks/dev/t001_stuff_c34.png
1
masks/dev/t001_things_bag.png
38
t002
[ 104288, 421144, 104800, 421656 ]
EPSG:28992
1
masks/dev/t002_stuff_c34.png
3
masks/dev/t002_things_bag.png
5
t003
[ 202712, 510168, 203224, 510680 ]
EPSG:28992
1
masks/dev/t003_stuff_c34.png
1
masks/dev/t003_things_bag.png
24
t004
[ 112744, 477768, 113256, 478280 ]
EPSG:28992
1
masks/dev/t004_stuff_c34.png
1
masks/dev/t004_things_bag.png
4
t005
[ 179656, 582640, 180168, 583152 ]
EPSG:28992
1
masks/dev/t005_stuff_c34.png
1
masks/dev/t005_things_bag.png
2
t006
[ 60824, 384224, 61336, 384736 ]
EPSG:28992
1
masks/dev/t006_stuff_c34.png
2
masks/dev/t006_things_bag.png
8
t007
[ 199016, 318472, 199528, 318984 ]
EPSG:28992
1
masks/dev/t007_stuff_c34.png
2
masks/dev/t007_things_bag.png
21
t008
[ 241128, 577224, 241640, 577736 ]
EPSG:28992
1
masks/dev/t008_stuff_c34.png
1
masks/dev/t008_things_bag.png
2
t009
[ 248136, 575168, 248648, 575680 ]
EPSG:28992
1
masks/dev/t009_stuff_c34.png
1
masks/dev/t009_things_bag.png
6
t010
[ 139424, 484544, 139936, 485056 ]
EPSG:28992
1
masks/dev/t010_stuff_c34.png
1
masks/dev/t010_things_bag.png
1
t011
[ 114168, 547088, 114680, 547600 ]
EPSG:28992
1
masks/dev/t011_stuff_c34.png
2
masks/dev/t011_things_bag.png
29
t012
[ 197360, 542712, 197872, 543224 ]
EPSG:28992
1
masks/dev/t012_stuff_c34.png
1
masks/dev/t012_things_bag.png
2
t013
[ 242184, 478680, 242696, 479192 ]
EPSG:28992
1
masks/dev/t013_stuff_c34.png
1
masks/dev/t013_things_bag.png
4
t014
[ 260568, 532904, 261080, 533416 ]
EPSG:28992
1
masks/dev/t014_stuff_c34.png
1
masks/dev/t014_things_bag.png
55
t015
[ 236904, 579248, 237416, 579760 ]
EPSG:28992
1
masks/dev/t015_stuff_c34.png
1
masks/dev/t015_things_bag.png
7
t016
[ 107480, 509784, 107992, 510296 ]
EPSG:28992
1
masks/dev/t016_stuff_c34.png
1
masks/dev/t016_things_bag.png
27
t017
[ 149032, 486984, 149544, 487496 ]
EPSG:28992
1
masks/dev/t017_stuff_c34.png
2
masks/dev/t017_things_bag.png
12
t018
[ 117992, 521512, 118504, 522024 ]
EPSG:28992
1
masks/dev/t018_stuff_c34.png
1
masks/dev/t018_things_bag.png
8
t019
[ 131752, 549240, 132264, 549752 ]
EPSG:28992
1
masks/dev/t019_stuff_c34.png
2
masks/dev/t019_things_bag.png
3

NL aerial proposal benchmark (200 tiles)

200 tiles of Dutch aerial imagery, each 512 × 512 px at 1 m (512 × 512 m), with instance masks for two kinds of object:

  • construction land, CBS NBBG2022 class 34 Bouwterrein ("stuff": amorphous, defined partly by use). 291 objects.
  • buildings, BAG panden ("things": crisp footprints, ≥ 200 m²). 3189 objects.

It was built for experiment 02 of my Posts series, which compares unsupervised mask proposals from diffusion self-attention against Segment Anything. The metric used there is COCO-style average recall at K proposals per tile.

How the tiles were chosen

Tiles are centred on randomly drawn NBBG2022 class-34 polygons of 1–20 ha anywhere in the Netherlands (seed 24), with a random offset of up to ±120 m and no two tiles overlapping. Every tile contains at least one construction object, kept only if ≥ 5000 m² of it lies inside the tile. Splits: dev = the first 20 tiles (for choosing hyperparameters), test = the other 180.

Files

  • dev/tNNN.png, test/tNNN.png: RGB, PDOK 2022_orthoHR (8 cm orthophoto served at 1 m).
  • masks/<split>/tNNN_stuff_c34.png, masks/<split>/tNNN_things_bag.png: 16-bit instance masks (0 = background, 1…N = objects). Paths are in the metadata columns stuff_c34_mask, things_bag_mask.
  • <split>/metadata.jsonl: tile ID, bounding box in RD New (EPSG:28992), object counts.

Rebuild from source with build_tiles.py in the repository.

Licences and attribution

  • Imagery: © Beeldmateriaal Nederland, via PDOK, CC BY 4.0
  • Construction land: CBS, NBBG2022 publicatiebestand, CC BY 4.0
  • Buildings: Kadaster, BAG via PDOK, CC0
  • This compilation: CC BY 4.0, Athithya Loganathan

Caveats

Class 34 includes serviced but un-built plots ("braak"), which look like grass. So part of the construction ground truth isn't visible in the image at all, by definition. Building masks are cadastral footprints, not roof outlines, so they can be offset from what the image shows.

Downloads last month
115