Introduction

The Athena API is a high performance API for interacting with the Resolver Unknown CSAM Detection Service. Documentation of the GRPC API itself along with the meaning of the underlying classification responses and API best practices can be found in the Athena Protobufs Documentation.

Classification can be performed in two ways:

• Synchronous: A single image is sent to the API and a response is returned immediately. This is suitable for low volume use cases, cases where latency is more valuable than throughput, or for testing and evaluation. Users wishing to perform synchronous classification should use the ClassifySingleImageAsync method on the AthenaClient class.
• Asynchronous: A bidirectional stream is opened to the API and images and responses are sent across that. Multiple clients are implemented to suit different use cases:
  • IAsyncEnumerable Client: A high level client that takes an IAsyncEnumerable and produces an IAsyncEnumerable of results. This client is suitable for most use cases and is the easiest to use. This client is implemented in the AthenaClient class via the ClassifyAsync method.
  • Low Level Channel-Based Client: A low level client that takes in a channel for sending images and returns a channel for receiving results. This client is suitable for users who need more control over the sending and receiving of images and results. This client is implemented in the AthenaApiClient class via the ClassifyAsync method.
  • TPL Dataflow Client: A high-level client that uses TPL Dataflow blocks to manage the sending and receiving of images and results. This client is suitable for users who are familiar with TPL Dataflow and want to use it to manage their classification pipeline. This client is implemented in the AthenaDataflowClient class. This client is packaged separately in the Resolver.Athena.DataflowClient NuGet package.