Create a Dataset

Press the "New Dataset" button, either on the top right or on the last card on the dataset page.

You will be guided through a 4-step wizard to set up your dataset. Stage 1 is to set a title. You can skip stages 2-4 with the "Continue Later" button if you are already familiar with the process.

Stage 2 will allow you to upload data. If you have less than 2,000 images to work with, feel free to drag and drop them. If instead you are working with more, or several (10+) heavy videos, we strongly recommend using the CLI commands provided. You can find more info on uploading via the API here.

You can also read about supported filetypes. We strive to support as many imaging modalities as the computer vision community needs, so if you believe your file format of choice isn't supported, and you'd like to see if natively imported, please let us know.

📘

Upload Limits

Internet browsers aren't made to upload many files at once. To ensure that your uploads don't fail half way, try using the CLI.
Through the browser, we recommend limiting yourself to the following batches:

• 2,000 images if under 5mb
• 1,000 images if under 10mb
• 25 videos if under 1,000 frames (30 seconds at 30fps).

Uploads will still work at higher quantities, but we can't guarantee a stable performance across all browsers.

In stage 3 you may define the classes and instructions for your dataset. You can edit these at any time in the future, so don't worry about adding all the info you need now.
Classes are the objects names you want to give to the items you wish to identify. They can be represented as a number of annotation types (bounding boxes, polygons, tags, etc.).

Instructions will be presented at the start of any task sent to users or workers. You may edit these as you go, and encounter edge cases to include as instructional material.

In stage 4 you will be prompted to configure the dataset's workflow. Check our our full guide here for instructions on setting up your team's workflow.