A new research paper identifies object-to-image (O2I) ratio mismatch as a primary cause of failure in visual instance retrieval, rather than resolution loss. The study, conducted on a benchmark of 3,021 Objaverse objects, found that over 80% of degradation in retrieval accuracy across different distances was due to O2I mismatch for most pretrained backbones. The research proposes solutions such as query-side scale augmentation and an OWLv2 crop reranker, which achieved state-of-the-art results on the ILIAS 100M dataset, demonstrating that O2I robustness can be learned. AI
IMPACT This research could improve the accuracy and robustness of visual search and retrieval systems by addressing a fundamental limitation.
RANK_REASON The cluster contains a research paper published on arXiv detailing a new finding and methodology in computer vision.
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
