How axelera-ai-hub/voyager-sdk works
Voyager SDK is an edge-inference SDK for deploying models to Axelera Metis devices and running them in application pipelines. The main user journey is model deployment, then configuring and executing an inference pipeline that accepts frames and produces model metadata/results. The repository also exposes Python pipeline composition, evaluation, display, and lower-level C++/GStreamer integration. The graph combines README-described capabilities with relationships supported by the supplied source index; unsampled implementation details are not treated as inspected.
Main entry points: network.py Network builder · manager.py Pipe manager · pipeline.py Pipeline translator · deploy.py Deploy CLI
Components 19
Model deployment
- Deploy CLI
deploy.py - Model compiler
axelera/app/compile.py
Pipeline runtime
- SDK configuration
axelera/app/config.py - Network builder
axelera/app/network.py - Pipeline translator
axelera/app/pipeline.py - Pipe manager
axelera/app/pipe/manager.py - Torch pipeline
axelera/app/pipe/torch.py - Device manager
axelera/app/device_manager.py
Inference operators
- Inference operators
axelera/app/operators/inference.py - Custom preprocessing
axelera/app/operators/custom_preprocessing.py - Multi-object tracking
trackers/src/TrackerFactory.cpp
Streaming APIs
- InferenceStream API
axelera/app/inf_tracers.py - AxInferenceNet
operators/axstreamer/src/AxInferenceNet.cpp - GStreamer plugins
operators/gstaxstreamer/src/GstAxPlugins.cpp
Data and results
- Dataset adapters
ax_datasets/objdataadapter.py - Evaluation
ax_evaluators/obj_eval.py - Inference metadata
axelera/app/meta - Result display
axelera/app/display.py
Other components
- Application developer
Connections 18
- Application developer → Deploy CLI: deploys model
README.md - Deploy CLI → Model compiler: requests compilation
- Application developer → Pipeline translator: configures pipeline
README.md - Pipeline translator → SDK configuration: reads settings
axelera/app/pipeline.py - Network builder → Model compiler: invokes compilation
axelera/app/network.py - Network builder → Inference operators: uses operators
axelera/app/network.py - Network builder → Pipeline translator: builds pipeline
axelera/app/network.py - Pipe manager → Network builder: loads network
axelera/app/pipe/manager.py - Pipe manager → Device manager: selects device
axelera/app/pipe/manager.py - Pipe manager → InferenceStream API: configures tracing
axelera/app/pipe/manager.py - Torch pipeline → Inference metadata: creates metadata
axelera/app/pipe/torch.py - Inference operators → Custom preprocessing: uses preprocessing
axelera/app/operators/inference.py - Dataset adapters → Evaluation: supplies evaluation data
ax_datasets/objdataadapter.py - InferenceStream API → Result display: renders results
axelera/app/inf_tracers.py - Application developer → AxInferenceNet: integrates inference
README.md - Application developer → GStreamer plugins: integrates streaming
README.md - AxInferenceNet → Inference metadata: transfers metadata
operators/axstreamer/src/AxInferenceNet.cpp - Application developer → Multi-object tracking: selects tracker
README.md