All changes to annotations and objects are tracked so that your dataset can be rolled back to any point-in-time. No more storing each version as a separate zip file.
Just like a code review, every change to an annotation is marked pending before it is accepted or rejected by an owner. This helps catch mistakes early on and prevents abuse.
Each action is signed for dataset integrity. When you generate a dataset release it will be fully reproducible. This means you don't have to worry about whether the release you found has been poisoned.
Annotations conform to schemas stored alongside the objects. No more deciphering XML or cleaning up inconsistent fields. These schemas are versioned so they can evolve with your dataset while still supporting historical releases. You can get creative and define your own!
Our storage is agnostic to your file type. Of course you can do text and image datasets, but if you have an idea for an exotic format, we'll version that too :) 3D models? PDFs? Audio? Let's go!
@ONLINE {, author = "Compute Heavy Industries Incorporated", title = "MekaBytes", year = "2022", url = "https://mekabytes.com", }