分布外检测中的隐形大猩猩效应
The Invisible Gorilla Effect in Out-of-distribution Detection
对做 OOD 检测的人，这篇论文提示：在近分布 OOD 场景下，检测分数会被伪影与 ROI 的视觉相似性（如颜色）系统性抬高或压低，因此设计检测器时应把伪影颜色/纹理与 ROI 的相似度作为控制变量，并用颜色交换的反事实样本做压力测试，而不是只在单一伪影类型上报告 AUROC。
原文依据 we identify a previously unreported bias in OOD detection
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论深度神经网络通过从图像中的感兴趣区域（ROI）学习特征，在视觉任务中取得了很高的性能，但当其被部署到与训练数据不同的分布外（OOD）数据上时，性能会下降。这一挑战催生了旨在识别并拒绝不可靠预测的 OOD 检测方法。尽管已有工作表明 OOD 检测性能会随伪影类型而变化，但其根本原因仍缺乏充分探索。为此，我们发现了 OOD 检测中一种此前未被报道的偏置：对于难以检测的伪影（近分布 OOD），当伪影与模型的 ROI 具有视觉相似性（例如颜色）时，检测性能通常会提升；而当二者不相似时，性能则会下降——我们将这一现象称为隐形大猩猩效应。例如，在一个以红色病灶为 ROI 的皮肤病变分类器中，我们表明 Mahalanobis Score 方法在检测 OOD 红色墨水（与 ROI 相似）标注时，其 AUROC 比检测黑色墨水（与 ROI 不相似）标注时高出 31.5%。我们对来自三个公开数据集（如 ISIC）的 11,355 张图像按颜色标注了伪影，并生成了颜色交换的反事实样本，以排除数据集偏置。随后我们在 7 个基准上评估了 40 种 OOD 方法，发现当伪影与 ROI 不同时，大多数方法的性能都出现了显著下降。我们的发现凸显了 OOD 检测中一个被忽视的失效模式，并为构建更稳健的检测器提供了指导。代码与标注见 https://github.com/HarryAnthony/Invisible_Gorilla_Effect。
英文原文摘要 1518 字符
Deep Neural Networks achieve high performance in vision tasks by learning features from regions of interest (ROI) within images, but their performance degrades when deployed on out-of-distribution (OOD) data that differs from training data. This challenge has led to OOD detection methods that aim to identify and reject unreliable predictions. Although prior work shows that OOD detection performance varies by artefact type, the underlying causes remain underexplored. To this end, we identify a previously unreported bias in OOD detection: for hard-to-detect artefacts (near-OOD), detection performance typically improves when the artefact shares visual similarity (e.g. colour) with the model's ROI and drops when it does not - a phenomenon we term the Invisible Gorilla Effect. For example, in a skin lesion classifier with red lesion ROI, we show the method Mahalanobis Score achieves a 31.5% higher AUROC when detecting OOD red ink (similar to ROI) compared to black ink (dissimilar) annotations. We annotated artefacts by colour in 11,355 images from three public datasets (e.g. ISIC) and generated colour-swapped counterfactuals to rule out dataset bias. We then evaluated 40 OOD methods across 7 benchmarks and found significant performance drops for most methods when artefacts differed from the ROI. Our findings highlight an overlooked failure mode in OOD detection and provide guidance for more robust detectors. Code and annotations available at https://github.com/HarryAnthony/Invisible_Gorilla_Effect.