weecology
feat(train): auto-log 5 annotated training samples at train start for annotation verification
Overview
Added automatic logging of a few annotated training images before training starts.
This helps verify image–annotation alignment and improves user debugging experience.
Implementation
- Added logic inside
on_train_start()incallbacks.py - Logs up to 5 sample annotations using
visualize.plot_annotations - Compatible with TensorBoard, WandB, and Comet loggers
- Human-readable code, minimal and maintainable
- No unrelated file formatting included
Testing
- Verified locally with multiple datasets
- Works with all supported loggers
Thanks for the contribution @Amitjoiya. Before we can consider the contents of the PR we need you to submit a clean PR that only includes the relevant changes, not large numbers of formatting changes across dozens of files. Once you've updated the PR accordingly let us know and we'll come back and consider the contribution.
@ethanwhite, Thanks for the review! I’ve cleaned up the PR to include only the relevant changes for the auto-logging feature. No formatting or unrelated modifications are present now. Please have another look when convenient 🙏
@ethanwhite, Please see my PR description now that's clean and meaningful. Thanks !
Sorry for the confusion. The main issue is with the code changes themselves. If you look at https://github.com/weecology/DeepForest/pull/1203/files you'll see that 29 files have been changed and almost all of the changes are formatting changes. That makes it too time consuming to read the code changes, so each PR should only include the changes to the code necessary to implement the specific thing you're working on. Does that make sense?
If you look at the Files Changed tab you'll see that you have still changed the formatting on (now) 31 files. I'm guessing you have some sort of automated system that is running Ruff on the entire code base (including files that are ignored by our build system). You need to turn it off so that you can submit a clean PR.