Failure building ML model from data set (only 2 of 8 batches get generated)

[RoboticsMB]

Hello,
I'm a member of team 2656, and we're running into a problem using Axon to build a model using a public dataset from the 2022 game, which I'm hoping you can help with.

Specifically, we tried to train a model using the dataset from team 1477
(https://www.chiefdelphi.com/t/1000-images-of-2022s-game-balls-for-vision/401125).
After downloading the dataset, we attempted to train the model using the standard presets:

• Epochs: 200
• Batch Size: 16
• Evaluation Frequency: 25
• Percent Evaluation: 30

When we run the training, Axon claims that finishs the work, but it only performs 50 of the designated 200 epochs.

This is highly reproduceable, and is something that we also see with other data sets (including some we've built and labeled ourselves), so we're fairly sure that it's not a 1-off thing.
(Note: in some cases, Axon doesn't produce any epochs, which we think may be related to issue #293 (#293)).

As additional background, we are running a Processor 11th Gen Intel(R) Core(TM) i7 @2.8GHz with 12Gb Ram and
64 bit Windows 11. We are using Docker version 4.6.1 and Axon version 0.1.0.

Attatched below are some screenshots of this scenario.

We're obviously very interested in using Axon and the models it can produce to improve our robot's performance. We would therefore greatly appreciate your help in resolving this issue.

Thank you,
Matthew Bandos
Team 2656