In theory, we could build an image for each pipeline step, or a subset of them (#3 ), but this may be cumbersome from an infrastructure point of view. More simply, perhaps, we could allow the build of a Docker image that would be intended for just a subset of the pipeline steps, allowing the builder to omit a subset of the (bulkier) dependencies like TensorFlow / tensorrt (#47 / #235 ) or Cellpose (e.g., if not running nuclei detection, or if running with the simpler threshold method).
In theory, we could build an image for each pipeline step, or a subset of them (#3 ), but this may be cumbersome from an infrastructure point of view. More simply, perhaps, we could allow the build of a Docker image that would be intended for just a subset of the pipeline steps, allowing the builder to omit a subset of the (bulkier) dependencies like TensorFlow / tensorrt (#47 / #235 ) or Cellpose (e.g., if not running nuclei detection, or if running with the simpler threshold method).