A new paper details the extensive process of re-annotating the ImageNet-1k dataset, revealing significant issues with the original labels. The researchers found that approximately 12% of labels were incorrect, 33.3% of images were multilabel, and 3.8% contained no relevant objects. This comprehensive re-annotation effort, named ReImageNet, includes multilabel correction, object localization, and revised class definitions, leading to accuracy improvements of up to 1.2% for supervised models and 5-6% for MLLMs. The study suggests that large-scale annotation requires iterative refinement and highlights the effectiveness of human-LLM collaboration for achieving high-quality results. AI
IMPACT Highlights the critical need for high-quality datasets in AI development and the potential for improved model performance with accurate data.
RANK_REASON The cluster contains an academic paper detailing a new dataset and methodology. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
