探测轻量视觉语言模型中视觉概念以用于自动驾驶
Probing Visual Concepts in Lightweight Vision-Language Models for Automated Driving
AI总结：
本研究探讨了轻量视觉语言模型在自动驾驶中的视觉概念编码问题，发现某些概念显式编码而其他隐式编码，并识别出感知和认知两种失败模式。
摘要：
视觉语言模型（VLMs）在自动驾驶应用中的使用正变得越来越普遍，其目的是利用它们的推理和泛化能力来处理长尾场景。然而，这些模型在处理与自动驾驶高度相关的简单视觉问题时往往表现不佳，而这些失败的原因仍不明确。在本工作中，我们检查了VLMs的中间激活，并评估了特定视觉概念在多大程度上被线性编码，目的是识别视觉信息流动中的瓶颈。具体来说，我们创建了仅在特定视觉概念上有所不同的反事实图像集，然后使用四个最先进的（SOTA）VLMs的激活来训练线性探测器来区分它们。我们的结果表明，诸如场景中物体或代理的存在等概念是显式且线性编码的，而其他空间视觉概念，如物体或代理的方向，则仅通过视觉编码器保留的空间结构隐式编码。同时，我们发现，在某些情况下，即使一个概念在模型的激活中被线性编码，模型仍无法正确回答问题。这导致我们识别出两种失败模式。第一种是感知失败，即回答问题所需的视觉信息未在模型的激活中线性编码。第二种是认知失败，即视觉信息存在，但模型未能将其正确对齐到语言语义。最后，我们展示了增加所涉对象的距离会迅速降低相应视觉概念的线性可分离性。总体而言，我们的发现提高了我们对VLMs在与自动驾驶高度相关的简单视觉任务中失败情况的理解。
英文摘要：
The use of Vision-Language Models (VLMs) in automated driving applications is becoming increasingly common, with the aim of leveraging their reasoning and generalisation capabilities to handle long-tail scenarios. However, these models often fail on simple visual questions that are highly relevant to automated driving, and the reasons behind these failures remain poorly understood. In this work, we examine the intermediate activations of VLMs and assess the extent to which specific visual concepts are linearly encoded, with the goal of identifying bottlenecks in the flow of visual information. Specifically, we create counterfactual image sets that differ only in a targeted visual concept and then train linear probes to distinguish between them using the activations of five state-of-the-art (SOTA) VLMs, including two training variants for one model. Our results show that concepts such as the presence of an object or agent in a scene are explicitly and linearly encoded, whereas other spatial visual concepts, such as the orientation of an object or agent, are only implicitly encoded by the spatial structure retained by the vision encoder, or are not linearly encoded at all. In parallel, we observe that in certain cases, even when a concept is linearly encoded in the model's activations, the model still fails to answer correctly. This leads us to identify two failure modes. The first is perceptual failure, where the visual information required to answer a question is not linearly encoded in the model's activations. The second is cognitive failure, where the visual information is present but the model fails to align it correctly with language semantics. Finally, we show that increasing the distance of the object in question quickly degrades the linear separability of the corresponding visual concept...
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
