arXiv:2605.29691v1 Announce Type: new 
Abstract: Self-supervised learning (SSL) has produced a diverse landscape of vision transformers (ViTs) whose pretrained representations support a wide range of downstream tasks. Towards a better understanding of these models, a body of work has assessed the mechanics of their self-attention as well as the types of information captured across their representations, revealing, for example, stark differences between models trained with contrastive learning (CL) and masked image modeling (MIM). However, these advances in model understanding have not yet fully permeated the broader community, where insights specific to CL models are sometimes generalized to MIM models. To make model understanding straightforward and intuitive for a broad audience, we propose a simple and easily interpretable visualization protocol.
  Our protocol is based on visualizing unsupervised semantic segmentation results, yet our goal is not to maximize segmentation performance. Instead, it allows us to convey model behaviors that consistently emerge across images. Benchmarking a diverse set of SSL models across layers and representations, we obtain novel insights into distinct positional biases and scaling behaviors, including strong boundary artifacts in DINOv3-Large model tokens. These insights complement and help communicate a range of previous findings.
  Our protocol further enables a clear visual distinction between positional effects and the closely related but distinct locality bias, which has been studied much more extensively in the literature. The protocol is publicly available on GitHub and we believe it will catalyze further model understanding for a broad community.

دراسة حديثة نُشرت على arXiv تقدم بروتوكولًا للتصور يهدف إلى تحسين فهم نماذج التعلم الذاتي، مع التركيز بشكل خاص على تقسيم المعاني غير المراقب. تسعى هذه الطريقة إلى توضيح الفروقات في سلوك النماذج، خاصة بين تلك المدربة باستخدام التعلم التبايني ونمذجة الصور المقنعة، دون إعطاء الأولوية لأداء التقسيم.

Un estudio reciente publicado en arXiv presenta un protocolo de visualización destinado a mejorar la comprensión de los modelos de aprendizaje auto-supervisado, centrándose especialmente en la segmentación semántica no supervisada. Este enfoque busca aclarar las diferencias en el comportamiento del modelo, especialmente entre aquellos entrenados con aprendizaje contrastivo y modelado de imágenes enmascaradas, sin priorizar el rendimiento de la segmentación.

Une étude récente publiée sur arXiv présente un protocole de visualisation visant à améliorer la compréhension des modèles d'apprentissage auto-supervisé, en se concentrant particulièrement sur la segmentation sémantique non supervisée. Cette approche cherche à clarifier les différences de comportement des modèles, notamment entre ceux formés avec un apprentissage contrastif et une modélisation d'image masquée, sans prioriser la performance de segmentation.

A recent study published on arXiv introduces a visualization protocol aimed at enhancing the understanding of self-supervised learning models, particularly focusing on unsupervised semantic segmentation. This approach seeks to clarify the differences in model behavior, especially between those trained with contrastive learning and masked image modeling, without prioritizing segmentation performance.