Collect performance metrics on HAB_pier:v001 using 20class model

[ktl014]

Story: As an ML engineer, I want to assess my model's performance on unseen data.

Subtasks

• [ ] Annotate test set (1 day of pier data)
• [ ] Use bash script to deliver annotations to svcl server
• [ ] Run model on test set
• [ ] Conduct failure analysis on model predictions

DoD: Accuracy, confusion matrix on test set + dataset size statistics

[ktl014]

Thoughts for later:

• Need to pull all data that sarah's annotated
• Need to clean up labels from the pier camera
• Run model on the test set, then upload predictions back to the website. Observe errors