dataset-validation

A minimal tool to generate and validate datasets.

Robot data integrity, before training.

Check image-text SFT data before training: broken records, exact train/eval overlap, and document-disjoint splits.

Veridex is the trust certificate for physical-AI data that tells you, across any format, whether the robot or sensor dataset you're about to train on is clean, correctly synchronized, and traceable to its source, before it silently wrecks your model.

Offline computer vision dataset auditing CLI tool for validating image datasets before model training.

YoloLint is a tool for automatic validation of dataset structure, annotation files, and image sizes in YOLO projects. It helps you catch typical errors in directory structure, YAML files, annotation files, and now also ensures all your images have the correct size before you start model training.

Pinned semantic-integrity audit and source-backed repair slice for LeRobot DROID

Refine distributed robot datasets into training-ready data — acceptance, episode-level filtering, deterministic cleaning, provenance (LeRobot v2/v3)

Image augmentation and YOLO annotation-validation utilities for computer-vision datasets.

Vet your robot datasets — 37 checks, repair & scoring for LeRobot data. Know it's UNSAFE TO TRAIN before wasting the run. Vets hf:// repos from 82 KB.

Automated audit, perceptual duplicate detection, and train/test leakage diagnostics for computer vision datasets.

All-in-one quality control for LabelMe and YOLO datasets: find broken files, label mistakes, bad shapes and split leakage before training, then fix them in one desktop app with full undo.

Reproducible quality audits of open robot-learning datasets — run on real files, evidence included.

Read-only PNG segmentation dataset checks: missing pairs, invalid labels, exact cross-split duplicates, and offline HTML reports. CPU-only Colab demo.

Browser tool that validates JSON datasets for AI training: missing fields, duplicate IDs, class distribution and a quality report.

dataset validity comparison tool

Python workbench for image annotation QA, train/test splits, and overlay reports

Validate dataset structure, relationships, temporal constraints, and integrity before release.

Read-only integrity checks for LeRobot v2.1 datasets, with command-line reports and CI exit codes.

Find the bad demonstrations before you train on them.