search

angelolab / Nimbus

Other
12 stars 1 forks source link

Wandb #80

Closed JLrumberger closed 7 months ago

JLrumberger commented 7 months ago

What is the purpose of this PR?

In this PR we add a residual unet with mirror padding as a new backbone architecture and wandb logging of validation metrics to the training pipeline.

How did you implement your changes I replaced the tensorboard logging functionality in ModelBuilder and PromixNaive with wandb logging and changed the vanilla unet architecture to use resnet style skip connections in the conv blocks. In addition I threw out a couple of tests from the deepcell PanopticNet, because they threw assertions without me changing any code or dependencies and I couldn't resolve them within a couple of hours.

Remaining issues

None

review-notebook-app[bot] commented 7 months ago

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

JLrumberger commented 7 months ago

Looks good! This is the one that had best performance in the benchmarking?

Yep