Training a Custom OD Model

### Product Category

JumpStart, Models, Pipelines, Quickstart Solutions

### Question

I have a project where I have used labelimg to label various bounding boxes for an object detection model. I have spent quite a bit of time now trying to install dependencies in order to train a model from TF model zoo on the new dataset, and think I am close by using a Docker TF image.

Before this: I tried to use Sagemaker thinking it could take my training data and then build a model for me so that I did not have to go through this process. I think I am close, but the training command and/or pipeline.config are not aligning leading to an issue where even though the checkpoint of the model is there: the training command is saying it is not. So, hoping to get through that step.

Still, this leads me to wonder if Sagemaker is supposed to assist with this: I had more difficulties with it than with installing and creating an app on my local machine. I have labels maps at the ready, and build the various tfrecords. Seems like it should have been able to take my labeling data, the model I want to base mine upon: and then run with it.

### Other Details

_No response_

Start with the Docker TensorFlow image, the training command, and pipeline.config, then verify how the checkpoint path and TFRecords are supplied. Compare that local setup with SageMaker's custom-training workflow; done means the expected checkpoint is found and a custom object-detection model can be trained from the prepared labels and data.

Written by the indexing model from the issue text.