CLI Reference¶

DatasetLint installs one console script:

datasetlint --help
datasetlint --version

The implementation uses one Typer entry point. lint, report, stats, diff, adapters, inspect, validate, and export-manifest are positional command words. The shorthand datasetlint DATASET_PATH still validates a native folder dataset.

Validate¶

datasetlint lint DATASET_PATH
datasetlint DATASET_PATH

Purpose: validate a native folder dataset.

Common options:

Option	Values	Purpose
--format, -f	console, json, markdown, html	Select output format
--fail-on	error, warning, info	Choose which severity makes the command exit non-zero
--config	path	Load config from a specific file
--checks	comma-separated groups	Run only selected check groups
--adapter	folder, auto, or adapter name	Select adapter

Examples:

datasetlint lint examples/minimal_dataset
datasetlint examples/minimal_dataset
datasetlint examples/minimal_dataset --checks labels,sync
datasetlint examples/minimal_dataset --format json
datasetlint examples/minimal_dataset --format html > report.html
datasetlint examples/minimal_dataset --fail-on warning

Report File¶

datasetlint report DATASET_PATH --out datasetlint-report.json
datasetlint report DATASET_PATH --out datasetlint-report.md
datasetlint report DATASET_PATH --out datasetlint-report.html

Purpose: write JSON, Markdown, or static HTML validation reports to files for CI artifacts, release evidence, or downstream tooling.

Stats¶

datasetlint stats DATASET_PATH
datasetlint stats DATASET_PATH --format json
datasetlint stats DATASET_PATH --format markdown

Purpose: compute dataset distributions without changing validation semantics.

Stats include sensors, frame counts, inferred rates, label class counts, confidence summary, track length summary, speed summary, acceleration summary, missing frame counts, and issue summary.

Diff¶

datasetlint diff OLD_DATASET NEW_DATASET
datasetlint diff OLD_DATASET NEW_DATASET --format json
datasetlint diff OLD_DATASET NEW_DATASET --fail-on-regression

Purpose: compare two folder datasets and classify changes, regressions, and improvements.

--fail-on-regression exits with code 1 when regressions are present.

Adapters¶

datasetlint adapters list
datasetlint adapters detect DATASET_PATH
datasetlint adapters DATASET_PATH
datasetlint adapters DATASET_PATH --format json
datasetlint adapters DATASET_PATH --format markdown

Purpose: list adapter availability or report which adapters detect a dataset path.

Inspect Adapter Manifest¶

datasetlint inspect DATASET_PATH --adapter coco
datasetlint inspect DATASET_PATH --auto-detect
datasetlint inspect hf://namespace/dataset --adapter huggingface --split train --deep --max-rows 1000
datasetlint inspect DATASET_PATH --adapter waymo --deep --max-rows 1000
datasetlint inspect DATASET_PATH --format json

Purpose: load a normalized DatasetManifest and print dataset name, adapter, sequence count, frame count, sensor streams, annotation count, splits, limitations, and warnings.

Validate Adapter Manifest¶

datasetlint validate DATASET_PATH --adapter kitti
datasetlint validate DATASET_PATH --auto-detect
datasetlint validate DATASET_PATH --adapter mcap --deep
datasetlint validate hf://namespace/dataset --adapter huggingface --split train --deep --max-rows 1000
datasetlint validate DATASET_PATH --format json

Purpose: run adapter-specific validation and return errors, warnings, coverage, validation mode, checked scope, unchecked scope, limitations, and stats. This is separate from datasetlint lint, which runs the native folder rule engine. Validation output includes coverage.common_rule_inputs and stats.common_rule_stats when decoded adapter records can feed shared manifest checks. Use --deep with nuScenes, MCAP, ROS bag, Waymo, or Hugging Face when you want bounded payload summaries, parser-backed channel/topic/frame metadata, or sampled Hugging Face row diagnostics instead of only file/cache indexing. If the requested deep parser is missing or cannot parse the input, validation reports valid: false and exits with code 1; parser warnings are not treated as successful deep validation.

Export Manifest¶

datasetlint export-manifest DATASET_PATH --adapter nuscenes --output manifest.json
datasetlint export-manifest DATASET_PATH --auto-detect --output manifest.json
datasetlint export-manifest DATASET_PATH --adapter rosbag --deep --output manifest.json

Purpose: write the normalized DatasetManifest JSON to a file.

Check Groups¶

--checks accepts:

• files
• metadata
• timestamps
• sensors
• sync
• calibration
• labels
• trajectories
• all

Multiple groups can be combined:

datasetlint examples/minimal_dataset --checks labels,sync

Exit Codes¶

Code	Meaning
0	Command completed and did not meet the configured failure threshold
1	Validation failed the --fail-on threshold, adapter validation failed, or diff regressions were found with --fail-on-regression
2	Invalid usage, unknown check group, bad adapter, or invalid config

Output Location¶

Validation output prints to stdout by default. Use datasetlint report --out when you want a JSON artifact, or redirect formatted output when you want a captured console or Markdown file:

datasetlint report examples/minimal_dataset --out datasetlint-report.json
datasetlint report examples/minimal_dataset --out datasetlint-report.html
datasetlint examples/minimal_dataset --format json > datasetlint-report.json
datasetlint examples/minimal_dataset --format markdown > datasetlint-report.md
datasetlint examples/minimal_dataset --format html > datasetlint-report.html