KALOS: Evaluate the quality of computer vision datasets
Systematic quality evaluation suite for AI/ML datasets. 103 ego datasets audited. ISO 5259-2 aligned.
Official repository for paper "Enhancing 3D Point Cloud Classification with ModelNet-R and Point-SkipNet"
面向研究、竞赛与论文场景的可追溯数据采集与交付工具
A Python toolkit for cleaner datasets in computer vision.
A Python library and CLI for dataset validation, schema checks, and basic drift signals.
Safe, auditable YOLO missing-label recovery | 安全可审计的 YOLO 漏标恢复工具
Validate OpenCV checkerboard camera-calibration datasets for blur, coverage, pose diversity, duplicates, and reprojection error.
Evaluation QA harness for misinformation datasets: stress tests evidence quality, shortcuts, ambiguity, and ranking fragility.
Local-first AI-assisted object detection annotation and quality-control desktop app
(WIP): 'Aporia' in Greek means 'inconsistent'. A Python library that detects and fixes dataset issues using both rule-based methods and ML models. It evaluates dataset quality across multiple metrics, including missing values, duplicates, outliers, class imbalance, and label consistency. It also suggests fixes based on the metric scores.
Industrial computer vision workflow for welding defect inspection using YOLO, OpenCV preprocessing, dataset QA, threshold governance, and edge-readiness analysis.
Error bars on a dataset vendor's quality claim: pre-registered measurement of Build AI's Egocentric-10K/100K hand-visibility and manipulation figures, with human gold, PPI intervals, and a distilled judge.
A health check for your fine-tuning datasets — diagnoses diversity, balance, and size risks before you waste a training run.
GenProof detects model collapse risk in pre-training datasets before training begins. It combines semantic entropy, tail-density, and AI detection into a composite probability score (ICS). Built with FastAPI and scikit-learn to help ensure data quality and compliance.
Time-aware dataset forensics and early-warning risk prediction for LLM fine-tuning. Predicts whether a dataset will damage a model — before the fine-tuning damage becomes visible.
The Dataset Quality Scoring Engine (DQS) evaluates the quality of any dataset using automated, model-agnostic metrics. The system processes user-uploaded datasets, computes embeddings, analyzes statistical and semantic properties, and outputs a standardized quality score
Find annotation rules your dataset never wrote down. Train a probe on a dataset's own labels, then measure where it disagrees.
Offline prompt and eval dataset linting for JSONL/CSV quality gates, PII, duplicates, split leakage, reports, and CI.
Read-only health checks and visual annotation reports for YOLO datasets.
To associate your repository with the dataset-quality topic, visit your repo's landing page and select "manage topics."