A 5-year solo side project with one mission
Be the fastest, most researcher-friendly place on the internet to find and download a labeled image dataset. Free forever.
Hi — I'm a computer vision engineer.
I've been building images.cv on the side for five years. It started because every new project ran into the same wall: hunt for a dataset, fix the folder layout, convert annotation formats, resize, split, re-download. I wanted one place that just works.
Today the catalog is 500+ labeled classes with millions of images, and roughly 400 researchers download something on any given day — mostly grad students, ML classes, and small teams prototyping a model. If that's you, you're the whole audience.
Because it's a one-person project, the roadmap is pragmatic: the catalog grows, downloads stay fast, features get added when they actually help someone ship a model. Everything else gets cut.
Two things keep the lights on
No ads on the catalog. No paywalls. No email-for-download trick.
Voluntary donations
Buy Me a Coffee. Every contribution goes straight to hosting and new datasets.
Custom B2B datasets
Companies that need specific coverage pay me to build it. The public catalog stays free.
What I care about
Research-friendly defaults
Clean folder layouts, reproducible outputs, BibTeX on every dataset, Colab starter notebooks. The details that actually matter for academic work.
Responsible use
Original licenses and attribution are respected. If a rights-holder ever asks for a takedown, the dataset comes down — no fight.
Open tooling, no lock-in
YOLO / COCO / Pascal VOC. Works with every common training framework. You own the data.