Search code examples
pythontensorflowobject-detectionobject-detection-api

How to run tensorflow object detection api (model_main_tf2) from code?

I want to run a training with the tensorflow object detection api. In the command line I can use the

python model_main_tf2.py --pipeline_config_path=path/to/pipeline.config --model_dir=path/to/trainedModel

But how can I start it from code?

model_main_tf2.FLAGS.pipeline_config_path = pipeline_config_path
model_main_tf2.FLAGS.model_dir = model_path
tf.compat.v1.app.run(model_main_tf2.main)

This is working, but the tf.compat.v1.app.run() calls the main function by using sys.exit(main). However I do not want, that the system exits.

How can I solve this?

Or maybe, how can I get around the usage of model_main_tf2 and the app.run()?

Solution

  • Since sys.exit(...) basically raises SystemExit you may just catch the SystemExit:

    try:
  tf.compat.v1.app.run(model_main_tf2.main)
except SystemExit:
  pass

# cool stuff happening here!