ABACUS: Adapting Unified Foundation Model for Bridging Image Count Understanding and Generation

  • 类型：arxiv
  • 标识：2606.23835
  • 链接：https://arxiv.org/abs/2606.23835
  • 主分类：multimodal
  • 形态：benchmark
  • 被引：0
  • 被引来源：Semantic Scholar + OpenAlex
  • S2被引：0
  • OpenAlex被引：0
  • 影响力被引：0
  • TLDR：ABACUS introduces three contributions: density-aware adaptive zooming paired with an objectness map from multi-head self-attention decomposition to spatially ground count predictions; a boundary-aware count policy trained via GRPO with nested local, boundary, and global rewards to eliminate over- and undercounting at crop boundaries.
  • OpenAlex ID：W7165728866
  • OpenAlex DOI：10.48550/arxiv.2606.23835
  • DOI：10.48550/arxiv.2606.23835
  • DOI来源：OpenAlex
  • 开放获取：green
  • 开放获取链接：https://doi.org/10.48550/arxiv.2606.23835
  • OpenAlex更新：2026-07-19
  • 待LLM分类：否
  • 标题中文：ABACUS：适配统一基础模型以桥接图像计数理解与生成
  • TLDR中文：ABACUS 提出三项贡献：与基于多头自注意力分解得到的目标性图相结合的密度感知自适应缩放，用于在空间上锚定计数预测；通过 GRPO 训练的边界感知计数策略，配合嵌套的局部、边界与全局奖励，以消除裁剪边界处的过度与不足计数。
  • 来源文件：
  • /inbox/tom/_candidates/2026-06-29-agent-memory-tool-use-candidates.json
  • /inbox/tom/_candidates/2026-06-28-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]