Gigadata is an image dataset collection and annotation platform. It allows anyone to easily contribute to the dataset by uploading images and annotating objects, and to use the dataset for training machine learning models.
The platform is designed to host a single huge dataset, which spans many classes, fields of interest, and use cases. Using querying it is easily possible to download only the parts that you need, though — for example, to get a JSON of the photos which contain either a cat or a dog (assuming these classes are registered on the server), and are under a PD equivalent licence:
want: - has: ["Domestic cat (Felis catus)", "Dog (Canis lupus familiaris)"] - nature: ["photo"] - licence: ["CC0-1.0", "X-public-domain", "X-informal-do-anything"]
Classes are hierarchical, so many search problems are solved. For example, consider this hierarchy (excuse my text art):
/- Aluminium food container
/- Aluminium household waste --- Aluminium can
/- Metal household waste /- Plastic bag
Household waste --- Plastic household waste --- Plastic bottle \
/ / \
Bottle ---------/---------------------------/ - PET bottle --- Clear PET bottle
/ / /
Plastic object ---- PET object --------------------------------/ /
\-------------------------- Clear PET object /
Multiple inheritance is also possible, seen here in Plastic bottle, for instance.
It's both a Plastic household waste and a Bottle, and because it's a Plastic
household waste, it's also a Household waste and a Plastic object. All sorts of
hierarchies like this one are possible; the has filter is used to search for
an object or its descendants.
There are more APIs, not just the search one. You can upload images, annotate them, and manage galleries programmatically.
If one wants more organisation of a certain set of images, a gallery feature is available where users can create galleries of images. Other users can also be assigned to add images to a gallery.
To prevent vandalism, you cannot change someone else's image annotations, but you can copy the image and make the changes; if the owner of the original approves, they can mark their version as obsolete and replaced by your version, which causes it to disappear from the search results.
The platform is made with Python, Flask and SQLAlchemy, just like the roundabout.
As always, this platform is free/libre under the AGPL. An official instance, Roundabout Datasets, is hosted at datasets.roundabout-host.com. Anyone can add images there, but they have to be free/libre. Nothing is guaranteed.
As far as I know, there's nothing else like this platform (at least not free/libre). I'd be happy to be proven wrong, though.
And why did I put "waste" in the topics? I'm moving the waste detection dataset there.