ABACUS：適應統一基礎模型以橋接圖像計數理解與生成

摘要

ABACUS 是一個統一的視覺語言模型，能處理物件計數、人群計數、指涉表達計數與計數忠實影像生成，且無需針對特定基準進行訓練。本模型以現有的 30 億參數統一基礎模型為基礎，並透過三項關鍵創新適應物件定位任務：利用物件性圖譜進行空間定位的密度感知自適應縮放；透過 GRPO 實現的邊界感知計數策略，以消除裁切邊界誤差；以及循環一致 GRPO 策略，讓理解分支自我批評生成的輸出，無需任何外部標註即可縮小理解與生成之間的落差。ABACUS 在七項基準測試中達到最先進的結果，不僅超越任務專屬的專家模型，也勝過規模更大的通用模型。

English

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.