Vision & documents, with a verdict on each.
Reading text off scans and photos, detecting and counting objects, and turning a camera feed into a number someone can act on.
Licensing bites hardest here — several of the best models are AGPL, which is a real cost if you ship them inside a product you sell.
Moving fastest this week: supervision (+1.3k), ultralytics (+977), MaaAssistantArknights (+638).
Adopt now
5Mature enough to put in production this quarter.
The connective tissue around vision models: drawing boxes, counting objects, tracking across frames, zone logic.
Instead of: A few hundred lines of fiddly, buggy annotation and counting code per project.
Cross-platform ML framework for media processing
Instead of: Manual ML model implementation
The standard library for image and video processing — cropping, detection, tracking, camera work.
Instead of: A pile of half-working image utilities copied from search results.
Data labeling and annotation tool
Instead of: Manual labeling or paid annotation tools
Large hub of ready-to-use datasets
Instead of: Manual dataset collection
Pilot it
3Worth a timeboxed spike before you bet on it.
Ready-to-run object detection and segmentation models you can train on your own images in an afternoon.
Instead of: A bespoke computer-vision contract, or a person watching a camera feed.
Object detection in PyTorch
Instead of: Manual image labeling
PyTorch Vision Transformer implementation
Instead of: manual CNN architecture design
Watch, or skip
4Real software; just not where a small team's next hundred hours should go.
Screen recording and AI agent integration tool
Instead of: Manual screen recording and annotation
Deep learning notes and code in Jupyter Notebooks
Instead of: Manual deep learning research and note-taking
OpenCV tutorials and examples
Instead of: Paid OpenCV courses or trial-and-error
Arknights game automation tool
Instead of: manual Arknights daily tasks
Vision & documents, in short
- What are the best open source computer vision and document AI right now?
- We would put supervision, mediapipe, opencv into production this quarter. supervision — The connective tissue around vision models: drawing boxes, counting objects, tracking across frames, zone logic. The full list below ranks 12 by momentum, size and how recently they shipped.
- Which of these can I use in a commercial product?
- 6 of the 12 carry a permissive licence (MIT, Apache-2.0, BSD). 6 do not — screenpipe (Custom / check the LICENSE file), ultralytics (AGPL-3.0), yolov5 (AGPL-3.0) and others. Copyleft and source-available licences carry obligations when you ship them inside something you sell. Read the LICENSE file; this is a reading of the label, not legal advice.
- How is this list ranked?
- By real week-over-week star movement first, with a size floor so an established tool is not buried by a two-week-old project, and a hard penalty for anything that has stopped shipping. Verdicts are an editorial call for a team of 2-20 with limited engineering hours — not a judgement on the software, and not the same call we would make for a lab.
Which of these matters to your business?
Tell us what you sell in one sentence and we will hand you three specific moves — priced per month, with the arithmetic shown. Free, no signup.
Give me three moves