AnnotateIt

Private, on-device computer vision annotation for images and video.

AnnotateIt is a local-first application for building computer vision datasets. Create projects, annotate images and video, review model-assisted proposals, check dataset quality, manage versions and splits, and export to standard formats.

Your media, labels and annotations stay on your device. Model files may be downloaded from the Hugging Face Hub, but inference runs locally on your hardware.

Open the web app · Website · Documentation · GitHub

AI-assisted computer vision annotation — running locally on your device.

Model artifacts

This organization is the official distribution point for model artifacts supported by AnnotateIt.

Every published model repository is expected to include:

• the upstream source, license and attribution;
• an exact preprocessing and class-mapping specification;
• a documented ONNX input/output contract;
• versioned files with SHA-256 checksums;
• PyTorch-to-ONNX Runtime validation results;
• known limitations and supported runtimes.

Available models

• EdgeCrafter ECDet-S — FP32 ONNX
• EdgeCrafter ECDet-M — FP32 ONNX

Both models are downloaded from immutable revisions, verified with SHA-256, and run locally in AnnotateIt.

Supported annotation workflows

• Object detection
• Instance segmentation
• Keypoint detection
• Image classification
• Semantic search and batch pre-labelling
• Dataset QA, versions, splits and standard-format export

AnnotateIt does not train models. It helps you build and validate datasets locally, then export them for training in the stack of your choice.