视觉几何基础 Transformer 中涌现的离群视图剔除能力
Emergent Outlier View Rejection in Visual Geometry Grounded Transformers
对做前馈式三维重建的人，这篇论文给出一个免训练、免监督的离群视图剔除手段：直接探测 VGGT 某一中间层的内部表征来打分过滤噪声输入，可即插即用地接到现有前馈重建流程的输入端。
原文依据 we identify a specific layer that naturally exhibits outlier-suppressing behavior
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论从野外图像集合中进行可靠的三维重建，常常受到噪声图像的干扰——这些无关输入与其他图像几乎没有或完全没有视角重叠。传统 Structure-from-Motion 流程通过几何验证与离群值剔除来处理这类情况，而前馈式三维重建模型缺乏这些显式机制，导致在野外条件下性能下降。本文发现，现有的前馈重建模型（如 VGGT）尽管没有显式的离群值剔除机制，也没有噪声感知训练，却能够内在地区分干扰图像。通过在不同比例合成干扰图像下的深入分析，我们定位到一个自然表现出离群抑制行为的特定层。进一步的探测表明，该层编码了具有判别性的内部表征，从而具备有效的噪声过滤能力；我们直接利用这一能力，在无需任何额外微调或监督的情况下，在前馈三维重建中执行离群视图剔除。在受控数据集与野外数据集上的大量实验表明，这种隐式过滤机制具有一致性，并能在多样场景中良好泛化。
英文原文摘要 1256 字符
Reliable 3D reconstruction from in-the-wild image collections is often hindered by noisy images--irrelevant inputs with little or no view overlap with others. While traditional Structure-from-Motion pipelines handle such cases through geometric verification and outlier rejection, feed-forward 3D reconstruction models lack these explicit mechanisms, leading to degraded performance under in-the-wild conditions. In this paper, we discover that the existing feed-forward reconstruction model, e.g., VGGT, despite lacking explicit outlier-rejection mechanisms or noise-aware training, can inherently distinguish distractor images. Through an in-depth analysis under varying proportions of synthetic distractors, we identify a specific layer that naturally exhibits outlier-suppressing behavior. Further probing reveals that this layer encodes discriminative internal representations that enable an effective noise-filtering capability, which we simply leverage to perform outlier-view rejection in feed-forward 3D reconstruction without any additional fine-tuning or supervision. Extensive experiments on both controlled and in-the-wild datasets demonstrate that this implicit filtering mechanism is consistent and generalizes well across diverse scenarios.