How document training and result of tensorflow object detection api?

0
votes

When I run some trainings with the Tensorflow Object Detection API, how can I document (save) the

  • batch size
  • number of epochs/steps
  • duration per step
  • model (ssd, faster r-cnn, ...)
  • backbone
  • pretrained (yes/no)
  • callback functions
  • loss-graphs
  • (Mean) Average Precision

Is there some tool to do this automatically or do I need to program it myself?

I have heard of Data Version Control or Apache Airflow, but I look for something simpler.

1
pythontensorflowkerasdeep-learningobject-detection-api

1 Answers

0
votes

For model training metadata like batch size, model type and loss function, it's usually recorded in json or yaml format.

For auto logging you can take a look at MLFlow for Tensorflow. There is also an official example.