Datasets:

sunnyallana
/

flood-depth-benchmark

image_id stringlengths 12 12	depth_bin stringclasses 7 values	reference stringclasses 14 values	note stringlengths 4 95	low_confidence bool 2 classes
flooded_2272	ankle	person	people walking in shallow water at far end; low confidence	true
flooded_1004	above_head	building	shacks flooded to roof line; boat	false
flooded_3179	unknown	null	flooded underpass, no reference object	false
flooded_0975	unknown	null	fast river torrent, nothing standing in water	false
flooded_2620	ankle	person	person standing in shallow street water	false
flooded_1203	knee	car	minivan in front of house, water at door bottom; low confidence	true
flooded_1116	knee	car wheel	cars driving, water at wheel hubs	false
flooded_2162	unknown	null	debris at bridge, people on bank only	false
flooded_0279	waist	door	water up shop doors along street; low confidence	true
flooded_3137	unknown	null	distant view of flooded town from bridge	false
flooded_3523	unknown	null	not a flood scene (orange blocks)	false
flooded_2766	unknown	null	distant river scene	false
flooded_0420	unknown	null	snowy park, shallow water at tree bases, no reference	false
flooded_1532	unknown	null	river in flood with ducks, no reference	false
flooded_0861	ankle	person	thin sheet of water on street, people at far end; low confidence	true
flooded_2443	dry	person	heavy rain on pavement, not flood depth	false
flooded_1530	unknown	null	river channel beside building	false
flooded_0635	unknown	null	distant trucks in flooded park, too far	false
flooded_1140	waist	person	two children pushing a cart, water at mid-thigh (~55 cm)	false
flooded_2571	knee	car wheel	car driving, water at wheel hubs	false
flooded_2395	unknown	null	flooded park path, no reference	false
flooded_3340	chest	car	SUV submerged to top of windows	false
flooded_0868	unknown	null	river against flood barrier	false
flooded_2506	unknown	null	flooded field and road, no reference	false
flooded_3632	ankle	door	shallow water at building doorsteps; low confidence	true
flooded_0666	unknown	null	flooded underpass, no usable reference	false
flooded_0030	unknown	null	river flooding park, distant	false
flooded_1471	unknown	null	aerial, buildings to roof line	false
flooded_0996	unknown	null	people on dry embankment, torrent depth unknown	false
flooded_3550	chest	building ground floor	water around ground-floor window line; low confidence	true
flooded_1517	knee	railing	riverside railings partly in water (non-standard reference)	false
flooded_1991	unknown	null	sandbag wall, river behind	false
flooded_2142	knee	bollard	bollards ~half submerged, crossing sign pole in water (non-standard reference)	false
flooded_2070	unknown	null	dam spillway	false
flooded_3630	unknown	null	aerial of flooded fields	false
flooded_2264	ankle	kerb	promenade with shallow water around trees; low confidence	true
flooded_1456	unknown	null	aerial, village flooded to roofs	false
flooded_1864	unknown	null	river in flood by fence	false
flooded_1938	dry	person	two people on wet pavement by sandbags, thin film of water	false
flooded_1263	unknown	null	wide shallow flood, sign pole only	false
flooded_2794	unknown	null	flooded street with billboard, no standard reference	false
flooded_0837	unknown	null	wide river, park	false
flooded_1689	unknown	null	flooded road at river, no reference in water	false
flooded_0129	unknown	null	floating debris	false
flooded_0572	unknown	null	people on bridge above torrent	false
flooded_1227	chest	car	cars on avenue submerged to window line	false
flooded_3669	ankle	person	restaurant car park, people near entrance in shallow water; low confidence	true
flooded_1139	knee	person	person wading beside car, water at knees / car door bottom	false
flooded_1494	unknown	null	flooded riverside street from stairs, lamp posts only	false
flooded_3316	unknown	null	aerial of flooded city	false
flooded_3116	unknown	null	no flood water visible (bridge, boats on trailers)	false
flooded_0732	unknown	null	river in flood between banks	false
flooded_1017	waist	person	rescue team with boat, man wading at thigh/waist	false
flooded_2987	unknown	null	river at bridge arches	false
flooded_2473	unknown	null	torrent beside buildings	false
flooded_0543	ankle	kerb	square and street in front of houses, shallow; low confidence	true
flooded_3214	unknown	null	weir	false
flooded_1564	knee	bench	park benches with water near seat level (non-standard reference)	false
flooded_2736	unknown	null	aerial	false
flooded_1029	unknown	null	aerial	false
flooded_0627	unknown	null	aerial of flooded suburb	false
flooded_0327	unknown	null	river at bridge	false
flooded_2886	unknown	null	channel with pipe	false
flooded_1169	chest	car	white car submerged to top of windows	false
flooded_1643	unknown	null	river with pontoon	false
flooded_1718	knee	fence	park fence about half submerged (non-standard reference); low confidence	true
flooded_2022	unknown	null	river over field	false
flooded_0964	knee	person	people with umbrellas wading, water at knees	false
flooded_0876	unknown	null	river with moored ships; people on dry promenade	false
flooded_3647	knee	person	elevated oblique view of alley, people wading mid-shin to knee	false
flooded_3017	unknown	null	black-and-white house, no flood water visible	false
flooded_0291	unknown	null	aerial	false
flooded_2334	unknown	null	river among trees	false
flooded_1953	ankle	door	shopfront with sandbags, shallow water at pavement; low confidence	true
flooded_1752	unknown	null	people on bridge steps, river below	false
flooded_0899	unknown	null	no-parking sign pole in water, no calibrated reference	false
flooded_0882	unknown	null	mostly dry road with barriers and sandbags	false
flooded_1379	knee	person	person beside cars, water at shins; low confidence	true
flooded_1082	unknown	null	aerial of flooded fields	false
flooded_0575	unknown	null	people on bridge above torrent	false
flooded_3014	unknown	null	weir torrent	false
flooded_1551	unknown	null	river between trees	false
flooded_1635	unknown	null	night street near underpass	false
flooded_0459	unknown	null	aerial	false
flooded_3650	unknown	null	aerial	false
flooded_2267	unknown	null	river under bridge	false
flooded_1466	unknown	null	people in a boat on open water	false
flooded_3149	unknown	null	flooded field near houses, grass showing, no reference	false
flooded_3607	unknown	null	receded: mud on road under railway bridge	false
flooded_0380	unknown	null	aerial	false
flooded_2173	unknown	null	underpass, sign posts only	false
flooded_1229	knee	car	street with cars, water at wheel hubs / door bottoms	false
flooded_0670	knee	fence	garden fence partly submerged (non-standard reference); low confidence	true
flooded_0027	knee	person	two children standing in water to mid-shin/knee	false
flooded_1850	unknown	null	flooded field like a lake	false
flooded_2386	knee	railing	steps into water by river, railing partly submerged; low confidence	true
flooded_3366	knee	railing	street railing about half submerged (non-standard reference); low confidence	true
flooded_2299	unknown	null	underpass, no water visible	false
flooded_2300	unknown	null	footbridge over torrent	false
flooded_2496	unknown	null	aerial, mud and destroyed bridge	false

End of preview. Expand in Data Studio

Flood Depth Benchmark

300 ground-level flood photos from the test split of the Close-View Flood Dataset (CVFD), each labelled with how deep the water is against an object of known height. It is the depth benchmark of Floodline, which estimates depth zero-shot from the waterline.

How the labels were made. Each image was labelled by visual inspection of contact sheets with an AI assistant (Claude, Anthropic) during the Floodline project, against the class definitions below. The labels have not yet been checked by a second, human annotator. Treat them as a first-pass reference, not as expert ground truth.

Class	Meaning	Images
`dry`	under 2 cm at the reference	16
`ankle`	2-20 cm	27
`knee`	20-50 cm	29
`waist`	50-100 cm	10
`chest`	100-160 cm	9
`above_head`	160 cm or more, or ground floor submerged	6
`unknown`	no reference object, or aerial / river view	203

reference names the object the depth was read against (person, car, door, building, ...). note is the labeller's free text; 42 rows say "low confidence", and low_confidence flags them.

Why two thirds are `unknown`

Most web-sourced flood imagery has nothing standing in the water that could calibrate depth: aerial views, rivers in flood, open water with no people, vehicles or doors in it. Those images are labelled unknown rather than guessed. A method that never says unknown on this set is overconfident.

Fields

Field	Type
`image`	image	224 × 224 RGB, EXIF removed
`image_id`	string	CVFD id, e.g. `flooded_2272`
`depth_bin`	string	one of `dry`, `ankle`, `knee`, `waist`, `chest`, `above_head`, `unknown`
`reference`	string	object the depth was read against, or empty
`note`	string	labeller's note
`low_confidence`	bool	the note says "low confidence"

Provenance and licence

Images come from CVFD (see its card): mostly FloodIMG (CC0) plus web-sourced scenes whose sources were not recorded. The labels are original work. Released under CC BY-SA 4.0. If an image is yours and you want it removed, open an issue at https://github.com/sunnyallana/floodline with the image id.

Citation

@dataset{shabanali2026flooddepth,
  author = {Shaban Ali, Sunny},
  title  = {Flood Depth Benchmark},
  year   = {2026},
  url    = {https://huggingface.co/datasets/sunnyallana/flood-depth-benchmark}
}

Downloads last month: 36

Flood Depth Benchmark

Why two thirds are unknown

Fields

Provenance and licence

Citation

Why two thirds are `unknown`