Skip to content

Use Your Own Model

This tutorial covers two different cases.

Case 1: Enable A Supported Public Wrapper

If the method is already supported in ST-CNVBench, you only need to configure and enable it.

  1. Start from configs/templates/models.template.yaml
  2. Copy the relevant model section into your own models.yaml
  3. Set enabled: true
  4. Fill in runtime-specific paths or image names
  5. Run with --models <ModelName>

See Model Run for the detailed model configuration contract.

Case 2: Add A New Wrapper

If your method is not yet supported, the public extension path is:

  1. Add a wrapper under src/st_cnvbench/model/tools_scripts/<model_name>/
  2. Register it in src/st_cnvbench/model/tools_scripts/__init__.py
  3. Add a public config section to configs/templates/models.template.yaml
  4. If the method should participate in evaluation, add or update the corresponding evaluation loader mappings

Before wiring a full execution path, make sure the new wrapper can at least:

  • locate the standardized dataset bundle
  • prepare its own input files
  • build a valid command line
  • fail loudly when required paths or files are missing

Try Next