---

name: detection-workflows description: "Guides SimpleDet config-driven training, bbox and mask evaluation, fine-tuning, inference-speed benchmarking, checkpoints, and GPU-aware workflow troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Detection workflows

Use this route for training, bbox evaluation, mask evaluation, inference speed, fine-tuning, FP16, model/config selection, or checkpoint output. Read cli-reference.md, then configuration.md before constructing a command.

Gate the workflow

SimpleDet is symbolic/static-shape MXNet. Its entry scripts import configs by turning config/foo.py into module config.foo, read roidbs from data/cache/, construct mx.gpu contexts from KvstoreParam.gpus, and write under experiments/. Confirm all three before any long run.

Use the bundled wrapper to make the checkout root explicit and avoid unsafe shell composition:

python <skill-root>/scripts/run_workflow.py --repo-root /path/to/simpledet \
  --entrypoint speed --config config/faster_r50v1_fpn_1x.py \
  --shape 800 1333 --gpu 0 --count 100 --dry-run

Remove --dry-run only after the setup diagnostic, config inspection, cache validation, checkpoint check, and required CUDA backend check pass. The wrapper does not make a workflow safe: the selected entry point can allocate GPUs, read large data, load weights, and write experiments.

Workflow routes

• train: fit or resume a detector; long-running and weight-mutating.
• test: bbox evaluation from a compatible checkpoint.
• mask-test: bbox plus instance-mask evaluation; requires polygon data.
• speed: static-shape one-GPU forward benchmark; requires --shape.

Use workflows.md for staged execution and troubleshooting.md for failures.

Route onward

• Missing runtime or CUDA/mxnext: setup-and-operations.
• Missing cache or bad annotations: data-preparation.
• New model or symbol contract: model-customization.

Use scripts/inspect_config.py for source/config inspection; it reports missing optional runtime packages rather than starting a symbol build. Use the root run_workflow.py wrapper for actual entry-point invocation.