AI & softwareResearch findings
Diagnosing Object Coverage Loss in Preference-Optimized Models and Restoring It via Base-Model Guidance
Original published 2026.11.18 03:24 KSTAdded to Blackdeck 2026.11.18 22:44 KST
What has changed?
By early stopping and appending high-confidence base-only categories, object coverage loss in preference-optimized models is restored without training
Key points
- Preference optimization reduces object coverage in vision-language models
- Early stopping and base-only category appending restore missing object categories without training
- Base-model-guided restoration achieves up to 86% of the coverage gain from optimized checkpoints
Connect it to your work
Preference optimization reduces object coverage; base-model-guided restoration recovers missing categories effectively
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.