arXiv:2511.16719v1 Announce Type: new 
Abstract: We present Segment Anything Model (SAM) 3, a unified model that detects, segments, and tracks objects in images and videos based on concept prompts, which we define as either short noun phrases (e.g., "yellow school bus"), image exemplars, or a combination of both. Promptable Concept Segmentation (PCS) takes such prompts and returns segmentation masks and unique identities for all matching object instances. To advance PCS, we build a scalable data engine that produces a high-quality dataset with 4M unique concept labels, including hard negatives, across images and videos. Our model consists of an image-level detector and a memory-based video tracker that share a single backbone. Recognition and localization are decoupled with a presence head, which boosts detection accuracy. SAM 3 doubles the accuracy of existing systems in both image and video PCS, and improves previous SAM capabilities on visual segmentation tasks. We open source SAM 3 along with our new Segment Anything with Concepts (SA-Co) benchmark for promptable concept segmentation.

تم تقديم نموذج Segment Anything Model (SAM) 3 كإطار موحد قادر على اكتشاف وتجزئة وتتبع الكائنات في الصور ومقاطع الفيديو باستخدام مطالبات مفاهيمية. يحسن هذا النموذج من Segmentable Concept Segmentation (PCS) من خلال استخدام محرك بيانات قابل للتوسع ينتج مجموعة بيانات تحتوي على 4 ملايين تسمية مفهوم فريدة، مما يحسن بشكل كبير من دقة التجزئة في الصور ومقاطع الفيديو.

Se ha presentado el Segment Anything Model (SAM) 3 como un marco unificado capaz de detectar, segmentar y rastrear objetos en imágenes y videos utilizando indicaciones conceptuales. Este modelo mejora la Segmentación Conceptual Prompable (PCS) al utilizar un motor de datos escalable que genera un conjunto de datos con 4 millones de etiquetas de conceptos únicos, mejorando significativamente la precisión de la segmentación en imágenes y videos.

Le Segment Anything Model (SAM) 3 a été introduit comme un cadre unifié capable de détecter, segmenter et suivre des objets dans des images et des vidéos en utilisant des invites conceptuelles. Ce modèle améliore la Segmentation Conceptuelle Prompable (PCS) en utilisant un moteur de données évolutif qui génère un ensemble de données avec 4 millions d'étiquettes de concepts uniques, améliorant considérablement la précision de segmentation dans les images et les vidéos.

The Segment Anything Model (SAM) 3 has been introduced as a unified framework capable of detecting, segmenting, and tracking objects in images and videos using concept prompts. This model enhances the Promptable Concept Segmentation (PCS) by utilizing a scalable data engine that generates a dataset with 4 million unique concept labels, significantly improving segmentation accuracy in both images and videos.

SAM 3: Segment Anything with Concepts

arXiv:2511.17201v1 Announce Type: new 
Abstract: In medical image segmentation, heterogeneous privacy policies across institutions often make joint training on pooled datasets infeasible, motivating continual image segmentation-learning from data streams without catastrophic forgetting. While the Segment Anything Model (SAM) offers strong zero-shot priors and has been widely fine-tuned across downstream tasks, its large parameter count and computational overhead challenge practical deployment. This paper demonstrates that the SAM paradigm is highly promising once its computational efficiency and performance can be balanced. To this end, we introduce the Alignment Layer, a lightweight, plug-and-play module which aligns encoder-decoder feature distributions to efficiently adapt SAM to specific medical images, improving accuracy while reducing computation. Building on SAM and the Alignment Layer, we then propose Continual Alignment for SAM (CA-SAM), a continual learning strategy that automatically adapts the appropriate Alignment Layer to mitigate catastrophic forgetting, while leveraging SAM's zero-shot priors to preserve strong performance on unseen medical datasets. Experimented across nine medical segmentation datasets under continual-learning scenario, CA-SAM achieves state-of-the-art performance. Our code, models and datasets will be released on \mbox{https://github.com/azzzzyo/Continual-Alignment-for-SAM.}

تقدم دراسة جديدة مفهوم التوافق المستمر لنموذج SAM (CA-SAM)، وهي استراتيجية تهدف إلى تحسين نموذج Segment Anything (SAM) لعمليات تقسيم الصور الطبية. تتناول هذه الطريقة التحديات الناتجة عن السياسات المتنوعة للخصوصية بين المؤسسات، والتي تعيق التدريب المشترك على مجموعات البيانات المجمعة، مما يسمح بالتعلم المستمر من تدفقات البيانات دون نسيان كارثي.

Un nuevo estudio presenta el Alineamiento Continuo para SAM (CA-SAM), una estrategia destinada a mejorar el Modelo Segment Anything (SAM) para la segmentación de imágenes médicas. Este enfoque aborda los desafíos de las políticas de privacidad heterogéneas entre instituciones que dificultan el entrenamiento conjunto en conjuntos de datos agrupados, permitiendo el aprendizaje continuo a partir de flujos de datos sin olvidar catastróficamente.

Une nouvelle étude présente l'Alignement Continu pour SAM (CA-SAM), une stratégie visant à améliorer le Segment Anything Model (SAM) pour la segmentation d'images médicales. Cette approche répond aux défis des politiques de confidentialité hétérogènes entre institutions qui entravent l'entraînement conjoint sur des ensembles de données regroupées, permettant un apprentissage continu à partir de flux de données sans oubli catastrophique.

A new study introduces Continual Alignment for SAM (CA-SAM), a strategy aimed at enhancing the Segment Anything Model (SAM) for medical image segmentation. This approach addresses the challenges of heterogeneous privacy policies across institutions that hinder joint training on pooled datasets, allowing for continual learning from data streams without catastrophic forgetting.

SAM 3: Segment Anything with Concepts

Was this article worth reading? Share it