Comment on page
Dataset formats are very different (coco, cityscapes and others), and there are many more modalities (pictures, videos and others), and even just with pictures, there are many variations of file formats.
Supervisely has three ways how to store your assets:
Store files locally
The default strategy is to place all uploaded and generated assets to the storage on the same server where Supervisely is installed, for example, on a hard drive.
Store files remotely
The instance administrator can configure the Supervisely platform to upload and store all generated assets to a remote cloud storage provider, such as S3. This can affect performance, since files need to be transferred over the network, but it is a more reliable method of data storage. This is only available on Enterprise Edition.
Store individual files remotely (“Import by Link”)
The hybrid approach that takes the best of both worlds. In this scenario, you store files locally (so that generated or uploaded manually files end up on a hard drive), but use special Supervisely Apps or API methods to add images or videos to datasets “by link”. That means, that instead of uploading the actual content of the file, you provide a resource url (such as “https://…” or “s3://…”) and Supervisely will not store the content on the platform — instead, it will load it from your remote provider on-fly. This method allows you to import huge datasets almost instantly, but you will have to manage the remote storage yourself.
Supervisely calculate file hashes when you upload your assets: because of that, we do not store duplicates.