ORIC：面向大型视觉语言模型上下文不协调下物体识别的基准测试
ORIC: Benchmarking Object Recognition under Contextual Incongruity in Large Vision-Language Models
ORIC 通过 LLM 引导采样和 CLIP 引导采样构造上下文不协调的物体-场景对，为评估 LVLM 的物体识别鲁棒性和幻觉提供了可复用的基准与训练数据构造流程；其视觉强化微调策略也可用于提升模型在非典型场景下的识别可靠性。
原文依据 we introduce the Object Recognition in Incongruous Context (ORIC) framework, which constructs incongruous object-context pairs through two complementary strategies
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论大型视觉语言模型（LVLM）通过结合视觉与语言，在图像描述、视觉问答和机器人领域表现出色，然而在非典型场景中，它们常常漏掉明显的物体或幻觉出不存在的物体。我们从不确定性的角度审视这些失败，聚焦于上下文不协调——即物体意外出现或未在预期上下文中出现——并表明此类情况增加了最先进 LVLM 的识别难度。为研究这一情形，我们引入了上下文不协调下的物体识别（ORIC）框架，该框架通过两种互补策略构建不协调的物体-上下文对：（1）LLM 引导的采样，用于识别图像中存在但难以识别的物体；（2）CLIP 引导的采样，用于挖掘看似合理但实际缺失的物体。将 ORIC 应用于 MSCOCO，我们创建了 ORIC-Bench 和 ORIC 风格的训练数据。对 18 个 LVLM 和 2 个开放词表检测器的评估揭示了在不协调上下文下显著的性能下降和偏差。在 600 个 ORIC 样本上对 Qwen3-VL-8B-Instruct 进行视觉强化微调，提升了其在 ORIC-Bench、AMBER 和 HallusionBench 上的性能。总体而言，我们表明上下文不协调是不确定性的一个关键来源，并为更可靠的 LVLM 提供了工具。数据集和代码已公开于 https://github.com/ZhaoyangLi-1/ORIC。
英文原文摘要 1348 字符
Large Vision-Language Models (LVLMs) excel at captioning, visual question answering, and robotics by combining vision and language, yet they often miss obvious objects or hallucinate nonexistent ones in atypical scenes. We examine these failures through the lens of uncertainty, focusing on contextual incongruity, where objects appear unexpectedly or fail to appear in expected contexts, and show that such cases increase recognition difficulty for state-of-the- art LVLMs. To study this regime, we introduce the Object Recognition in Incongruous Context (ORIC) framework, which constructs incongruous object-context pairs through two complementary strategies: (1) LLM-guided sampling to identify hard-to-recognize objects present in the image and (2) CLIP-guided sampling to mine plausible but absent ones. Applied to MSCOCO, ORIC creates ORIC-Bench and ORIC-style training data. Evaluating 18 LVLMs and 2 open-vocabulary detectors reveals significant degradation and bias under incongruous contexts. Visual Reinforcement Fine-Tuning of Qwen3-VL-8B-Instruct on 600 ORIC sam- ples improves performance on ORIC-Bench, AMBER, and HallusionBench. Overall, we show that contextual incongruity is a key source of uncertainty and provide tools for more reliable LVLMs. The dataset and code are publicly available at https://github.com/ZhaoyangLi-1/ORIC.