arXiv:2511.21367v1 Announce Type: new 
Abstract: Endoscopic (endo) video exhibits strong view-dependent effects such as specularities, wet reflections, and occlusions. Pure photometric supervision misaligns with geometry and triggers early geometric drift, where erroneous shapes are reinforced during densification and become hard to correct. We ask how to anchor geometry early for 4D Gaussian splatting (4DGS) while maintaining temporal consistency and efficiency in dynamic endoscopic scenes. Thus, we present Endo-G$^{2}$T, a geometry-guided and temporally aware training scheme for time-embedded 4DGS. First, geo-guided prior distillation converts confidence-gated monocular depth into supervision with scale-invariant depth and depth-gradient losses, using a warm-up-to-cap schedule to inject priors softly and avoid early overfitting. Second, a time-embedded Gaussian field represents dynamics in XYZT with a rotor-like rotation parameterization, yielding temporally coherent geometry with lightweight regularization that favors smooth motion and crisp opacity boundaries. Third, keyframe-constrained streaming improves efficiency and long-horizon stability through keyframe-focused optimization under a max-points budget, while non-keyframes advance with lightweight updates. Across EndoNeRF and StereoMIS-P1 datasets, Endo-G$^{2}$T achieves state-of-the-art results among monocular reconstruction baselines.

تم تقديم مخطط تدريب جديد يسمى Endo-G$^{2}$T لتحسين التشتت الغاوسي الرباعي الأبعاد (4DGS) لمشاهد الفيديو التنظيري. تتناول هذه الطريقة التحديات مثل التأثيرات المعتمدة على العرض والانحراف الهندسي من خلال دمج تقطير الأولويات الموجهة هندسيًا وحقول غاوسية مدمجة زمنياً لتحسين الاتساق الزمني.

Se ha introducido un nuevo esquema de entrenamiento llamado Endo-G$^{2}$T para mejorar el splatting gaussiano 4D (4DGS) en escenas de video endoscópico. Este método aborda desafíos como los efectos dependientes de la vista y la deriva geométrica al incorporar una destilación de prior guiada por la geometría y un campo gaussiano incrustado en el tiempo para mejorar la coherencia temporal.

Un nouveau schéma de formation appelé Endo-G$^{2}$T a été introduit pour améliorer le splatting gaussien 4D (4DGS) pour les scènes vidéo endoscopiques. Cette méthode aborde des défis tels que les effets dépendants de la vue et la dérive géométrique en intégrant une distillation de prior géométrique et un champ gaussien intégré dans le temps pour une meilleure cohérence temporelle.

A new training scheme called Endo-G$^{2}$T has been introduced to enhance 4D Gaussian splatting (4DGS) for endoscopic video scenes. This method addresses challenges such as view-dependent effects and geometric drift by incorporating geometry-guided prior distillation and a time-embedded Gaussian field for improved temporal consistency.

Endo-G$^{2}$T: Geometry-Guided & Temporally Aware Time-Embedded 4DGS For Endoscopic Scenes

arXiv:2512.07596v1 Announce Type: new 
Abstract: The recent Segment Anything Model (SAM) 3 has introduced significant advancements over its predecessor, SAM 2, particularly with the integration of language-based segmentation and enhanced 3D perception capabilities. SAM 3 supports zero-shot segmentation across a wide range of prompts, including point, bounding box, and language-based prompts, allowing for more flexible and intuitive interactions with the model. In this empirical evaluation, we assess the performance of SAM 3 in robot-assisted surgery, benchmarking its zero-shot segmentation with point and bounding box prompts and exploring its effectiveness in dynamic video tracking, alongside its newly introduced language prompt segmentation. While language prompts show potential, their performance in the surgical domain is currently suboptimal, highlighting the need for further domain-specific training. Additionally, we investigate SAM 3's 3D reconstruction abilities, demonstrating its capacity to process surgical scene data and reconstruct 3D anatomical structures from 2D images. Through comprehensive testing on the MICCAI EndoVis 2017 and EndoVis 2018 benchmarks, SAM 3 shows clear improvements over SAM and SAM 2 in both image and video segmentation under spatial prompts, while zero-shot evaluations on SCARED, StereoMIS, and EndoNeRF indicate strong monocular depth estimation and realistic 3D instrument reconstruction, yet also reveal remaining limitations in complex, highly dynamic surgical scenes.

تم تقديم نموذج Segment Anything (SAM) 3، الذي يظهر تقدمًا في قدرات التقسيم، والإدراك ثلاثي الأبعاد، وإعادة البناء في الجراحة الروبوتية. يدعم هذا النموذج تقسيم الصفر باستخدام مجموعة متنوعة من المطالبات، بما في ذلك المدخلات المعتمدة على اللغة، مما يعزز مرونة التفاعلات. تسلط التقييمات التجريبية الضوء على أدائه في تتبع الفيديو الديناميكي والحاجة إلى مزيد من التدريب في التطبيقات الجراحية.

Se ha presentado el modelo Segment Anything (SAM) 3, que muestra avances en las capacidades de segmentación, percepción 3D y reconstrucción en cirugía robótica. Este modelo admite segmentación de cero disparos utilizando varios prompts, incluidas entradas basadas en lenguaje, mejorando así la flexibilidad de las interacciones. Una evaluación empírica destaca su rendimiento en el seguimiento de video dinámico y la necesidad de un entrenamiento adicional en aplicaciones quirúrgicas.

Le modèle Segment Anything (SAM) 3 a été introduit, mettant en avant des avancées dans les capacités de segmentation, de perception 3D et de reconstruction en chirurgie robotique. Ce modèle prend en charge la segmentation zéro-shot à l'aide de divers prompts, y compris des entrées basées sur le langage, améliorant ainsi la flexibilité des interactions. Une évaluation empirique met en lumière ses performances dans le suivi vidéo dynamique et la nécessité d'un entraînement supplémentaire dans les applications chirurgicales.

The Segment Anything Model (SAM) 3 has been introduced, showcasing advancements in segmentation, 3D perception, and reconstruction capabilities in robotic surgery. This model supports zero-shot segmentation using various prompts, including language-based inputs, enhancing interaction flexibility. An empirical evaluation highlights its performance in dynamic video tracking and the need for further training in surgical applications.

Endo-G$^{2}$T: Geometry-Guided & Temporally Aware Time-Embedded 4DGS For Endoscopic Scenes

Was this article worth reading? Share it

MicroEstimates

3YOURMIND

4o Image Gen