Skip to main content

Algorithms and reports


Developers and data scientists can deploy algorithms with Tator, providing a convenient interface for initiating batch processing of media and for AI-assisted annotation. Once an algorithm is wrapped in an Argo Workflow and registered in Tator, users can launch thousands of asynchronous workloads on their media with one click. Both GPU and CPU-based algorithms are supported. Organization administrators have flexibility in where the workloads are processed; any Kubernetes cluster, local or cloud-based, can be used to process algorithms. Remote processing can be configured on a per-algorithm basis, allowing processing hardware configuration that is tailored to the algorithm. Lightweight Argo workflows can be used to interface with third party processing services, such as AWS SageMaker or GCP DataLab.

For AI-assisted annotation, Tator provides the ability to register custom front-end applets. Through custom applets, users can submit the current frame or a video clip around the current frame to standing model servers such as TensorFlow Serving or NVidia's Triton Inference Server, or they can perform client-side processing in the browser using WASM-based tools like OpenCV.js or TensorFlow.js.


Through the same mechanism as custom algorithms, developers can integrate custom report generation workflows into Tator. This provides a simple interface through which media can be filtered and selected using media attributes, folders, or other criteria, and then be submitted for inclusion in report generation inputs. As with algorithms, report generation is highly scalable. Once report generation completes, it can be hosted as a generic file in Tator, or, through the Tator REST API, workflows can send automated emails to project members that include report generation outputs as attachments.

Read more