Reproduce “Renovating Names in Open-Vocabulary Segmentation Benchmarks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haiwen Huang, Songyou Peng, Dan Zhang, Andreas Geiger
RENOVATE, an automatic framework, enhances the semantic precision of object names in open-vocabulary segmentation benchmarks by generating context-aware labels. The approach significantly improves open-vocabulary model performance, for example, achieving nearly 4 additional PQ points on MS COCO, and facilitates more accurate model evaluation by differentiating between semantically similar and truly incorrect predictions.
Key insights
- Systematically improving segment-level names in open-vocabulary segmentation benchmarks provides richer and more accurate textual supervision, directly enhancing model training and generalization capabilities.
- Renovated names significantly improve data efficiency, allowing models to achieve comparable or superior performance with substantially less training data than when using original names.
- High-quality, fine-grained ground-truth names enable a more nuanced evaluation of models by revealing "benign" misclassifications (semantically similar but textually different) and facilitating detailed per-category analysis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.