Datasets:

luuuulinnnn
/

nexar-collision-prediction

time_of_event float64	time_of_alert float64	light_conditions string	weather string	scene string	time_to_accident float64
21.324	20.216	Normal	Clear	Urban	0.5
18.385	15.446	Normal	Rain	Highway	0.5
19.789	17.928	Normal	Rain	Other	0.5
20.232	18.763	Normal	Clear	Urban	0.5
20	18.867	Normal	Cloudy	Urban	0.5
19.467	17.067	Normal	Clear	Urban	0.5
19.758	17.581	Normal	Clear	Highway	0.5
19.533	16.1	Normal	Clear	Highway	0.5
21.667	19.6	Normal	Clear	Sub-urban	0.5
18.367	16.467	Normal	Clear	Urban	0.5
19.167	18.233	Normal	Clear	Urban	0.5
19.8	18.624	Normal	Clear	Urban	0.5
19.784	18.054	Normal	Cloudy	Sub-urban	0.5
19.92	18.246	Normal	Cloudy	Sub-urban	0.5
20.187	17.05	Normal	Clear	Urban	0.5
19.767	17.8	Normal	Clear	Urban	0.5
19.967	17.611	Normal	Clear	Rural	0.5
19.2	17.9	Normal	Clear	Urban	0.5
20.16	18.853	Normal	Cloudy	Urban	0.5
22.129	21.452	Normal	Clear	Urban	0.5
20.717	19.65	Normal	Cloudy	Urban	0.5
20.134	17.743	Normal	Clear	Rural	0.5
19.6	18	Normal	Clear	Urban	0.5
19.58	17.837	Normal	Cloudy	Urban	0.5
19.8	17.933	Normal	Clear	Highway	0.5
19.533	15.6	Normal	Clear	Highway	0.5
19.067	18.633	Normal	Clear	Urban	0.5
19.967	18.4	Normal	Clear	Urban	0.5
20.654	19.286	Normal	Clear	Urban	0.5
20.7	20	Normal	Cloudy	Urban	0.5
19.948	19.612	Normal	Cloudy	Urban	0.5
20.033	18.033	Normal	Clear	Highway	0.5
20.483	18.033	Normal	Clear	Urban	0.5
20.7	19.733	Normal	Clear	Highway	0.5
19.633	18.6	Normal	Clear	Urban	0.5
19.4	17.867	Normal	Clear	Highway	0.5
19.267	19.133	Normal	Clear	Highway	0.5
14.147	12.175	Normal	Cloudy	Urban	0.5
19.904	17.681	Normal	Rain	Highway	0.5
19.1	15.767	Normal	Clear	Other	0.5
19.544	18.558	Normal	Clear	Urban	0.5
19.983	19.65	Normal	Cloudy	Urban	0.5
21.823	19.6	Normal	Clear	Highway	0.5
20.8	19.433	Normal	Cloudy	Urban	0.5
19.596	18.99	Normal	Cloudy	Urban	0.5
19.079	18.479	Normal	Clear	Rural	0.5
19.333	18.133	Normal	Cloudy	Urban	0.5
19.967	18	Normal	Clear	Urban	0.5
20.126	17.839	Normal	Clear	Urban	0.5
19.5	18.933	Normal	Cloudy	Sub-urban	0.5
20.667	17.267	Twilight	Clear	Highway	0.5
20.467	19.167	Normal	Cloudy	Urban	0.5
19.967	18.867	Normal	Clear	Urban	0.5
19.033	19	Normal	Clear	Urban	0.5
20.967	19.033	Normal	Rain	Urban	0.5
19.333	17.5	Normal	Clear	Urban	0.5
19.833	18.5	Normal	Clear	Urban	0.5
20.5	17.767	Normal	Clear	Highway	0.5
19.9	18.767	Normal	Clear	Urban	0.5
20.492	19.708	Normal	Cloudy	Urban	0.5
20.367	18.9	Normal	Clear	Urban	0.5
20.217	19.733	Normal	Cloudy	Urban	0.5
20.267	19.499	Normal	Rain	Urban	0.5
19.85	19.817	Normal	Clear	Urban	0.5
19.852	18.839	Normal	Clear	Urban	0.5
19.071	18.85	Normal	Clear	Urban	0.5
20.367	19.8	Normal	Cloudy	Urban	0.5
20.067	18.567	Normal	Rain	Highway	0.5
23.454	21.494	Normal	Clear	Urban	0.5
26.852	24.827	Normal	Cloudy	Urban	0.5
18.079	16.81	Normal	Clear	Urban	0.5
18.401	16.44	Twilight	Clear	Urban	0.5
18.47	17.041	Normal	Clear	Urban	0.5
20.943	19.048	Normal	Clear	Urban	0.5
18.958	17.679	Normal	Clear	Urban	0.5
19.267	18.467	Normal	Rain	Urban	0.5
20.133	19.067	Normal	Cloudy	Urban	0.5
21.021	19.253	Normal	Cloudy	Highway	0.5
20.913	19.752	Normal	Clear	Urban	0.5
21.202	19.327	Normal	Rain	Urban	0.5
21.667	19.167	Normal	Clear	Urban	0.5
11.033	8.033	Normal	Clear	Sub-urban	0.5
22.467	18.667	Normal	Cloudy	Sub-urban	0.5
18.981	16.923	Normal	Clear	Other	0.5
19.447	18.725	Normal	Rain	Urban	0.5
20.55	19.412	Normal	Cloudy	Urban	0.5
20.721	16.583	Twilight	Clear	Sub-urban	0.5
20.24	18.67	Normal	Cloudy	Highway	0.5
20.283	18.846	Normal	Clear	Highway	0.5
20.467	17.7	Normal	Clear	Urban	0.5
19.661	18.681	Normal	Cloudy	Highway	0.5
18.5	16.933	Normal	Cloudy	Highway	0.5
20.367	19.633	Normal	Clear	Sub-urban	0.5
16.912	15.476	Dark	Clear	Urban	0.5
19.233	16.5	Normal	Clear	Urban	0.5
21.233	17.233	Normal	Clear	Highway	0.5
20.133	18.233	Normal	Clear	Highway	0.5
23.619	20.128	Normal	Clear	Urban	0.5
9.628	8.029	Normal	Clear	Industrial	0.5
21.833	20.367	Normal	Cloudy	Urban	0.5

End of preview. Expand in Data Studio

Nexar Dashcam Collision Prediction — test splits

A mirror of the test-public and test-private splits of the Nexar dashcam collision-prediction dataset. This is not the original release — see Provenance and license below before using it.

The training split is not included here; only the two test splits are.

Split	Videos	Size
`test-public/positive`	334	1.4 GB
`test-public/negative`	332	1.4 GB
`test-private/positive`	338	1.5 GB
`test-private/negative`	339	1.4 GB
Total	1343	5.5 GB

positive = the clip contains a collision or near-collision event. negative = it does not.

Each split directory holds its .mp4 files plus a metadata.csv.

Note: the four metadata.csv files together list 1344 rows while 1343 .mp4 files are present, so one referenced file is missing from this mirror.

Video format

All clips are uniform:

1280 × 720, ~30 fps
10 s long (300 frames)
single forward-facing dashcam, H.264 in MP4

Roughly 3.7 hours / ~400k frames in total.

Metadata

metadata.csv columns:

Column	Meaning
`file_name`	clip filename within the split directory
`time_of_event`	timestamp (s) of the collision / near-collision
`time_of_alert`	timestamp (s) at which the event becomes foreseeable
`light_conditions`	Normal / Twilight / Dark / Bright
`weather`	Clear / Cloudy / Rain / Fog
`scene`	Urban / Highway / Sub-urban / Rural / Industrial / Other
`time_to_accident`	`time_of_event − time_of_alert`

For negative clips the event-timing fields are empty.

There is no camera calibration, no intrinsics or extrinsics, and no multi-camera data — this is monocular forward-view footage only.

Label distribution (all 1344 rows)

Lighting — Normal 1245 · Twilight 65 · Dark 26 · Bright 8 Weather — Clear 873 · Cloudy 399 · Rain 69 · Fog 3 Scene — Urban 690 · Highway 344 · Sub-urban 227 · Other 37 · Rural 32 · Industrial 14

The distribution is heavily skewed toward normal lighting and clear/cloudy weather. Treat Rain (69), Dark (26), Fog (3) and Industrial (14) as small-sample slices, not as balanced conditions.

Loading

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="luuuulinnnn/nexar-collision-prediction",
    repo_type="dataset",
)

import csv, os

split = f"{path}/nexar_collision_prediction/test-public/positive"
rows = list(csv.DictReader(open(f"{split}/metadata.csv")))
print(rows[0])
# {'file_name': '00002.mp4', 'time_of_event': '21.324', 'time_of_alert': '20.216',
#  'light_conditions': 'Normal', 'weather': 'Clear', 'scene': 'Urban',
#  'time_to_accident': '0.500'}

Provenance and license

The data originates from Nexar and was distributed for the Nexar dashcam collision-prediction challenge. This repository is a re-upload of the test splits only, made for convenience; no files were modified, re-encoded or re-labelled.

Two things to be aware of before you use or redistribute this:

The original terms govern. The competition's rules and the dataset licence — not this page — determine what you may do with these files. Redistribution of competition test splits is often restricted. Check the original terms and obtain the data from Nexar directly if you need a clean licensing position.
These are real road recordings. Faces, licence plates and locations appear in the footage and are not anonymised.

If you are the rights holder and want this mirror taken down, open a discussion on this repository.

Citation

Please cite the original Nexar dataset and challenge, not this mirror.

Downloads last month: 119