Anonymous Mode

Use Aquarium without directly sharing your raw data.

We built Anonymous Mode to allow customers to use Aquarium in cases where sharing the raw data outside their organization is undesirable or impossible.

When using Anonymous Mode, users provide access-controlled URLs for the imagery.

  • These URLs reference the image locations on customer-controlled infrastructure.

  • The raw data is not uploaded to Aquarium's servers.

When using Aquarium with Anonymous mode, the raw data is accessed and displayed only by the browser without ever leaving the customer's environment.

Note that when using Anonymous Mode, Aquarium cannot access the raw data and therefore cannot generate embeddings for you using our available pre-trained models.

If you want to use features powered by embeddings (Similarity Search, Collection Campaigns, the Embeddings View), you'll need to provide your own embeddings.

Aquarium's Data Model

Datasets in Aquarium are made up of:

  • Raw Data (imagery, pointclouds, etc.)

  • Labels (bounding boxes, masks, etc.)

  • Inferences (from your model on your dataset)

  • Metadata (time of collection, device ID, etc.) about the data/labels/inferences Aquarium should index.

  • Embeddings generated from a neural network for each raw data element, label and inference. By default Aquarium generates embeddings using a set of pre-trained models. Customers can optionally provide embeddings from their own models instead.

Aquarium uses the raw data itself for only two purposes:

  • Rendering a visualization in the user's browser via client-side code.

  • Generating an embedding vector via a single run through a pre-trained model.

Most insights in Aquarium are generated using the indexed metadata, the relationships between the labels and inferences, and the relationships between embedding vectors.

With Anonymous Mode configured, Aquarium cannot access the raw imagery and therefore cannot generate embeddings using our pre-trained models.

Customers can choose to provide their own embeddings during upload, which Aquarium will use in place of the default embeddings. Labels, Inferences and Metadata are still shared as normal.

As long as customers provide their own embeddings, choosing to use Anonymous Mode does not limit access to any of Aquarium's functionality.

Configuring Anonymous Mode

There are a number of ways to share data. In general we've found the simplest option to be hosting the data in a private bucket and configuring local credentials.

[Preferred] - Private Bucket and Local Credentials

Your data is in a private storage bucket and Aquarium will never access the raw data. This implementation works by providing bucket paths that point to secure resources.

When your users want to view the resources in Aquarium's web UI, they use local credentials to view the data. Neither the credentials, nor the data ever leaves your users' browser / local device.

Network Restricted URLs

Resources are restricted behind the corporate VPN. This implementation works by restricting the URLs to only being accessible from your company's corporate network.

When your users want to view the resources in Aquarium's web UI, they'll be able to do so only from the corporate network. The data is not accessible to Aquarium's servers.

Locally Hosted Files

Resources are hosted locally on your device, or on a network mounted storage device similar to your file system. This implementation works by hosting a simple local file server and making the imagery accessible via an HTTP request from the browser.

When your users want to view the resources in Aquarium's web UI, they'll be able to do so only from their device using URLs similar to http://localhost:5000/path/to/local/img.png. The data is not accessible to Aquarium's servers.

Last updated