---

name: architecture-configuration description: "Explains VAD's registered detector, head, transformer, dataset, coder, and loss components and guides safe edits to VAD configuration families." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

VAD architecture and configuration

Use this route for model internals, plugin registration, config inheritance, stage selection, model variants, or VADv2 integration questions.

Route

1. Read configuration-reference.md to select tiny/base and stage-1/stage-2/e2e.
2. Read model-and-plugin-api.md before changing a registered component or tensor contract.
3. Run python scripts/check_config_contract.py CONFIG --check-plugin for a safe config-only check. It does not build a model or access data.
4. Keep plugin=True and the plugin directory importable before OpenMMLab builders resolve custom registry names.
5. For data files and temporal annotations, use data-preparation; for launch/checkpoint behavior use training-evaluation.

Actual VAD model execution is CUDA/native-extension dependent. A config parse is not proof that the detector can be built.

Scope boundaries

• Data roots, converter, and CAN-bus: data-preparation.
• Train/eval CLI and released-weight normalization: training-evaluation.
• Prediction rendering: visualization.

See troubleshooting.md for registry, compiler, shape, and stage failures.