UniCounting: Instance-Aware Proposal Consolidation for Image-Query-Free Multi-Category Counting

Key points

• Visual counting is commonly formulated as counting a single specified target, with a model receiving an image-specific exemplar, text query, or target category and returning a single count.1
• We instead study fixed-vocabulary image-query-free multi-category counting.1
• A 3,267-parameter category-shared relation head predicts same-instance affinities from instance-mask-derived supervision.1
• On COCO clean500, UniCounting obtains lower point-estimate vector $\ell1$ error and absent-class false mass than calibrated OWLv2-All80, with comparable micro presence F1.1

Sources (1)

• [1]UniCounting: Instance-Aware Proposal Consolidation for Image-Query-Free Multi-Category Counting
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 02:02 PM

  We present UniCounting, which casts counting as instance-aware structural inference over an over-complete proposal set.

  Visual counting is commonly formulated as counting a single specified target, with a model receiving an image-specific exemplar, text query, or target category and returning a single count.

Extractive summary: sentences quoted from the sources.