跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 6 天前AI 評分49

ViRA:透過跨輸入相對策略最佳化,邁向超高解像度視覺智能體

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization

AI 導讀

研究團隊提出 CrossRPO,以對比集訓練 ViRA 視覺智能體,令模型須讀取超高解像度圖像中的證據才能答對。在最高解像度的 ViRA-Bench 上,CrossRPO 的整組準確率為 63.3%,高於 GRPO 的 59.3%。

正文

Published on Nov 16

Authors:

,

,

,

,

,

,

,

,

,

,

Abstract

Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, with CrossRPO on only about 3,500 items of ViRA-Train each. CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO. ViRA-9B is the most accurate open model of its size on all benchmarks, and ViRA-27B outperforms Claude Haiku 5.5 and GPT-6-Luna, both at their highest thinking level, on most benchmarks.

View arXiv page View PDF Project page GitHub Add to collection

Get this paper in your agent:

hf papers read 2611.21605

Don't have the latest CLI?

curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 2

huaweilin/ViRA-9B

Image-Text-to-Text •

9B •

Updated about 11 hours ago

•

72

huaweilin/ViRA-27B

Image-Text-to-Text •

27B •

Updated about 11 hours ago

•

71

Datasets citing this paper 2

huaweilin/ViRA-Bench

Viewer •

Updated about 11 hours ago

•

840

•

184

huaweilin/ViRA-Train

Preview •

Updated about 11 hours ago

•

96

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.21605 in a Space README.md to link it from this page.

Collections including this paper 1

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co