weecology / weecology/DeepForest

Can't we do better than destroy the entire classification head when the annotations don't match?

Open
#798 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Ideas for Machine Learning!
Dominant language
Python
Stars
774
Forks
265
PR merge metrics
No merged PRs in 30d

Description

I think we can provide users with more insight than just a random instantiation of the backbone when we are running multi-class and retraining to add new labels.

https://deepforest.readthedocs.io/en/latest/advanced_features/multi_species.html

We could grab the classification head from the release model. We could get just the top layers. I think this was one of the workarounds we identified and then didn't return to.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the multi-species retraining guidance at advanced_features/multi_species.html, then trace the code path used when annotations do not match during multi-class retraining. Determine the expected behavior for retaining useful portions of the release model, and define tests that show retraining no longer falls back to a random backbone in this case.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.