| title | ABACUS: Adapting Unified Foundation Model for Bridging Image Count Understanding and Generation |
|---|---|
| authors | Anindya Mondal, Sauradip Nag, Anjan Dutta |
| date | 2026-06-22 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2606.23835 |
| upvotes | 5 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | ABACUS is a unified vision-language model that handles object counting, crowd counting, referring-expression counting, and count-faithful image generation without any benchmark-specific training required. Our model is built on existing 3B-parameter unified foundation model and is adapted for object localization tasks using three key innovations: density-aware adaptive zooming with objectness maps for spatial grounding; a boundary-aware count policy via GRPO to eliminate crop-boundary errors; and a cycle-consistent GRPO strategy where the understanding branch self-critiques generated outputs, closing the understanding-generation gap without any external annotations. ABACUS achieves state-of-the-art results across seven benchmarks, outperforming both task-specific specialists and larger generalist models. |
cite
@misc{mondal2026abacus,
title={ABACUS: Adapting Unified Foundation Model for Bridging Image Count Understanding and Generation},
author={Anindya Mondal and Sauradip Nag and Anjan Dutta},
year={2026},
eprint={2606.23835},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2606.23835}
}