INDUS-AD—Indian Dataset of Unstructured Urban Scenes for Autonomous Driving—contains 85.31 recording-timeline hours across 1,469 driving recordings (approximately 4.006 TB) across Bengaluru (Bangalore), India. It brings together 9,927 face/plate-blurred multi-camera MP4s, 588.57 aggregate camera-view hours summed across cameras (not distinct driving hours), normalized sensor tables, and derived semantic and depth layers in one directly usable release.

Each recording is organized around a stable recording_id that joins camera video with camera timing and calibration, privacy-transformed GNSS/pose fields, frame-level camera pseudo-labels, and dense semantic/depth archives. The result supports non-commercial research in end-to-end driving, video and world models, multi-camera perception, localization, synchronization, and robotics data tooling.

Researchers define filtering, sampling, synchronization, targets, and partitions for their tasks. Synchronized steering, throttle, and brake targets are not provided.

Six-camera surround view from Bengaluru.

Key facts

Property	Value
Recordings / video Viewer rows	1,469
Camera MP4 files	9,927
MP4 payload bytes	3,956,842,789,114
Approximate published repository size	4.006 TB
Recording timeline (longest camera stream per recording)	85:18:25.609 (85.31 hours)
Aggregate camera-view duration	588:34:01.225 (588.57 hours)
Face/plate-blurred MP4 files	9,927
Named camera positions	7
GNSS recording tables	1,469
Camera-frame tables	9,927
Frame-label tables	1,469
Camera-annotation Viewer rows	34,306
Semantic/depth Viewer rows	34,306
GNSS Viewer rows	18,373,026
Dense-index tables	69
Dense depth archives	68
Dense semantic archives	1
Authored benchmark split	None

Comparison with major public driving datasets

The comparison separates released recording length, collected hours, and summed camera-hours. Gated and registered datasets are included.

Dataset	Geography/scope	Released data and recording length	Published size	Access/license	Comparison note
INDUS-AD	Bengaluru, India	1,469 recordings; 85.31 h released timeline; 588.57 summed camera-view h	About 4.006 TB encoded release	Manual gate; CC BY-NC 4.0	9,927 camera MP4s and normalized GNSS/calibration tables. Source and decoded representations overlap and must not be added.
NVIDIA PhysicalAI Autonomous Vehicles	Global, 25 countries	1,700 h; 306,152 20-second clips	133 TB	Gated; NVIDIA AV Dataset License	Multimodal clips, not India-focused.
OpenDV-2K / OpenDV-YouTube	Global, 244+ cities	More than 1,700 h in reproducible YouTube component	About 3 TB video; about 24 TB processed images	Source-reproducible; underlying source terms	Website video list; full-suite collection and reproducible component are distinct.
nuPlan	Four international cities	More than 1,300 h; 15,000+ logs	No single directly comparable total stated in cited overview	Registration; dataset terms	Planning logs and sensor packages have different scope.
BDD100K	United States	100,000 approximately 40-second videos; more than 1,000 h	Not stated in cited overview	Dataset terms	Independent videos; not India-focused.
Waymo Open Motion	United States	574 scenario-hours; 103,354 20-second segments	Not stated in cited overview	Dataset license	Scenario time, not continuous road-coverage time.
Argoverse 2 Sensor	United States	4.2 h; 1,000 approximately 15-second scenes	1 TB extracted	CC BY-NC-SA 4.0	Selected multimodal scenes.
IDD-3D	Hyderabad, India	Publisher reports over 5 h; 12,000 annotated LiDAR frames	236 GB	Project access; dataset terms	Camera/LiDAR corpus; annotated-frame count is separate from recording length.
ThirdEye Indian Road Dataset	Delhi NCR, India	645,714 annotated keyframes; released continuous video duration not stated	About 210 GB	CC BY 4.0	Image/mask/annotation TARs and GPS; continuous video is not listed.
METEOR	India	1,250 one-minute videos; approximately 20.83 h	Approximately 100 GB reported	Dataset terms	Annotated traffic videos with GPS trajectories and traffic-behaviour labels.
India Driving Dataset / IDD	India	117K-image portal release; continuous video duration not stated	72 GB for IDD 117K	Registration; research terms	Image-centric subsets; raw collection time is not a released-video duration.
DriveIndia / TiHAN camera release	India	66,986 annotated images reported; 120+ h describes collection, not verified released continuous footage	Not stated	EULA and access request	Official conference abstract separates selected images from 120+ collection hours.
TIAND / TiHAN multimodal	India	Download request offered for 2–4-minute scenes; complete released duration not stated	Author repository describes a roughly 37 GB sample	Usage agreement and approval	60 TB is advertised as collection size; complete released duration is unspecified.

Checked 16 October 2026. The India-focused duration ranking covers documented released recordings; unknown durations remain unranked. Storage sizes depend on encoding and representation.

Ways to explore INDUS-AD

INDUS-AD is an MP4-and-table dataset. Researchers can start in the Viewer, download selected camera videos, work directly with normalized sensor tables, or resolve full semantic and depth members through the dense indexes.

Dataset resources

The following supporting resources are packaged with this dataset:

Camera coverage

Conceptual layout; use the calibration tables for geometry. cam_front_tilted is available for 1,113 recordings.

Camera	Recordings with video	Missing recordings
`cam_front`	1,469	0
`cam_front_left`	1,469	0
`cam_front_right`	1,469	0
`cam_front_tilted`	1,113	356
`cam_rear`	1,469	0
`cam_rear_left`	1,469	0
`cam_rear_right`	1,469	0

Missing camera slots are null.

Bengaluru coverage

The recordings span Bengaluru (Bangalore). Brighter map areas show greater recording concentration.

Driving-condition examples

Selected frames illustrate clear, overcast, wet-road, and low-light conditions. Condition labels are machine-generated.

Researcher-controlled data policy

Select the published layers and define the sampling, synchronization, and partition policy for your research task.

Dataset structure

data/
  metadata.jsonl
  manifests/
    video-integrity.jsonl
    video-privacy-release.json
    quality-summary/
      camera_duration_summary.csv
      quality_summary_manifest.json
  viewer/
    camera-annotations/*.parquet
    gnss/*.parquet
    semantic-depth/*.parquet
    videos/*.parquet
  videos/
    <camera>/<two-hex-bucket>/<recording_id>.mp4
  tables/
    auxiliary-messages/
    calibrations/
    camera-frames/
    camera-messages/
    gap-events/
    gnss/
    keyframes/
    recording-metadata/
    recording-quality-summary/
    sync-groups/
  annotations/camera/
    frame-labels/sample_group=<group>/
    dense-index/
    depth/
    semantic/

The date-free recording_id is the stable join key across video, GNSS, camera timing, frame labels, and dense archive indexes. The two-character video bucket is sha256(recording_id)[:2]; it is only a deterministic storage partition. Use data/metadata.jsonl rather than guessing paths.

Annotation folders use sample_group=train, sample_group=validation, and sample_group=test. These and the legacy split column are sampling groups, not benchmark partitions. See the path migration map.

Dataset Viewer

The Viewer has four separate configurations so rows with different meanings are never flattened into one table:

videos (default): one row per recording with recording_id and seven camera video columns;
camera_annotations: one row per sampled camera frame with semantic, depth, weather, and image-quality pseudo-label fields;
gnss: one row per GNSS message with timing, privacy-transformed coordinates, motion, orientation, validity, and RTK/accuracy fields;
semantic_depth: one row per sampled camera frame with a colorized semantic-mask preview, a colorized metric-depth preview, exact join keys, and the per-frame depth range in meters.

Video checksums are available in data/manifests/video-integrity.jsonl.

Missing tilted-camera values remain null. Video paths in data/metadata.jsonl are relative to data/.

The Viewer’s technical train label is retained for API compatibility; researchers define their evaluation partitions.

Download one camera video

import json
from huggingface_hub import HfApi, hf_hub_download

repo_id = "gagandeepreehal/indus-ad"
revision = HfApi().repo_info(
    repo_id,
    repo_type="dataset",
    revision="main",
).sha
recording_id = "17-53-27.393-d655e06f8e7ee863e68443502137f2fe"

metadata_path = hf_hub_download(
    repo_id=repo_id,
    repo_type="dataset",
    revision=revision,
    filename="data/metadata.jsonl",
)

with open(metadata_path, encoding="utf-8") as stream:
    rows = (json.loads(line) for line in stream)
    row = next(row for row in rows if row["recording_id"] == recording_id)

relative_path = row["cam_front_video_file_name"]
video_path = hf_hub_download(
    repo_id=repo_id,
    repo_type="dataset",
    revision=revision,
    filename=f"data/{relative_path}",
)
print(video_path)

The example requires accepted gated access and a logged-in Hugging Face account. It resolves main once, then uses that immutable commit for both downloads. Replace recording_id with any value from data/metadata.jsonl. Download narrow file patterns rather than cloning the complete multi-terabyte repository.

Timing, joins, and GNSS privacy transformation

Recording IDs and paths are date-free. Tables retain time-of-day nanoseconds and recording-relative offsets for alignment.

The producer confirms hardware triggering and synchronization. Observed camera cadence is approximately 10 FPS; across 3,028,347 decoded camera groups, median timestamp spread is 0.020 ms and p95 is 0.025 ms. See the sensor reference for hardware and timing details.

The combined GNSS Viewer contains 15,317,896 gnss messages and 3,055,130 rec_gnss messages. For stream-specific rates and trajectories, use data/tables/gnss/ and filter its channel column. The flattened Viewer omits this field.

GNSS publication follows these rules:

precise local ENU coordinates remain fully precise;
global ENU and global latitude/longitude are published at approximately 1 m horizontal resolution;
height is retained;
sensor-quality and RTK diagnostics such as east/north sigma, position RMS, HDOP, and fix type are retained;
published_horizontal_resolution_m records the 1.0 m publication resolution;
rows with missing source coordinate objects remain null and invalid—coordinates are never invented; is_valid=false can also occur on rows that still contain sensor-reported coordinates.

Join by recording, camera identity, and compatible published timestamps, accounting for missing streams and gaps.

Camera pseudo-labels and dense archives

The frame-label tables contain model-generated weak pseudo-labels and summary features, not human ground truth. They include semantic class fractions, metric-depth summaries, weather probabilities, image-quality values, timing/join keys, and archive members. The dense index maps frame rows to full-resolution semantic or depth members stored in TAR archives.

The semantic_depth Viewer configuration provides 320×180 display previews for all 34,306 paired semantic masks and depth maps. The semantic_mask image column preserves class IDs as palette indices. The depth_map image column visualizes the normalized 16-bit map, while depth_min_m and depth_max_m give its corresponding metric range. These previews are for browsing—the authoritative 1280×720 uint8 semantic masks and uint16 depth maps remain in the dense TAR archives.

See the third-party notices for upstream attribution and licensing.

Annotation coverage and interpretation

The 1,469 source annotation tables contain 34,306 sampled rows, matching the Viewer count. At approximately 10 FPS, the 588.57 camera-view hours imply roughly 21.19 million frames: approximately 0.16% sampling density, or 3.46 sampled rows per camera stream on average. Exact per-frame and per-class completeness has not been measured. Use the dense index for frame/archive joins; schema details are available separately.

Intended uses

Appropriate non-commercial uses include end-to-end trajectory research with carefully derived pose-based future-motion targets; video foundation models and world models; camera perception and representation learning; temporal and cross-camera synchronization; GNSS/pose alignment; localization experiments; calibration; data-quality tooling; teaching; and research demonstrations.

Limitations and responsible use

Coverage and sensor availability vary by recording; frames, timestamps, and joins can contain gaps. Automated face/plate blur can miss identifying content, and trajectories retain linkage risk. The dataset is for research and has not been validated for production driving or safety decisions.

Do not use the data to re-identify or track individuals or expose sensitive locations.

Report sensitive or improperly included material privately through the Minus Zero contact form or team@minuszero.ai with subject Dataset privacy or takedown request. Include the repository and date-free recording_id or path when known. Do not post identifying evidence in a public discussion.

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). Attribution is required, commercial use is prohibited, and modifications must be indicated. The license does not remove privacy, publicity, trademark, or other third-party rights.

Citation

@dataset{indus_ad_2026,
  author    = {Gagandeep Reehal},
  title     = {INDUS-AD: Indian Dataset of Unstructured Urban Scenes for Autonomous Driving},
  year      = {2026},
  publisher = {Gagandeep Reehal},
  url       = {https://huggingface.co/datasets/gagandeepreehal/indus-ad}
}

Authorship and maintenance

Author and publisher: Gagandeep Reehal.

Use the repository discussions for non-sensitive questions and corrections. Use the private contact route above for privacy and takedown reports.

Downloads last month: 812

Total file size:

4.01 TB

Collection including gagandeepreehal/indus-ad

INDUS-AD: The largest Indian autonomous driving dataset

Collection

Subsample of datasets created by us to build foundational models for autonomous driving at Minus Zero • 4 items • Updated 14 days ago • 1