arXiv:2506.09782v2 Announce Type: replace 
Abstract: The Segment Anything Model 2 (SAM2) is a powerful foundation model for promptable segmentation. However, its high computational and memory costs are a major barrier to deployment on resource-constrained devices. In this paper, we present Q-SAM2, an accurate low-bit quantization method that achieves high compression and high fidelity. To address performance degradation arising from challenging weight and activation distributions during quantization, Q-SAM2 introduces two novel contributions: Variance-Reduced Calibration (VRC), an initialization method that reduces weight statistical variance by minimizing the Frobenius norm over a small calibration batch; and Learnable Statistical Clipping (LSC), a Quantization-Aware Training (QAT) method that learns momentum-stabilized clipping factors to manage outliers in weights and activations. Comprehensive experiments demonstrate that Q-SAM2 achieves highly accurate inference with substantial efficiency gains, significantly surpassing state-of-the-art general QAT schemes, particularly in the ultra-low 2-bit regime. Specifically, Q-SAM2 achieves an accuracy gain of up to 9.7 ppt in J&F on the video segmentation benchmark and 7.3 ppt in mIoU for instance segmentation over the best competing QAT model, all while achieving an 8x reduction in model size compared to the BF16 baseline.

تم تحسين نموذج Segment Anything Model 2 (SAM2) من خلال تقديم Q-SAM2، وهي طريقة لتقليل الكمية ذات البت المنخفض التي تقلل بشكل كبير من التكاليف الحاسوبية والذاكرة مع الحفاظ على دقة عالية. تعالج هذه الخطوة التحديات المتعلقة بنشر SAM2 على الأجهزة ذات الموارد المحدودة، مما يجعلها أكثر وصولاً لمجموعة متنوعة من التطبيقات.

El Segment Anything Model 2 (SAM2) ha sido mejorado con la introducción de Q-SAM2, un método de cuantificación de bajo bit que reduce significativamente los costos computacionales y de memoria mientras mantiene una alta fidelidad. Este avance aborda los desafíos de implementar SAM2 en dispositivos con recursos limitados, haciéndolo más accesible para diversas aplicaciones.

Le Segment Anything Model 2 (SAM2) a été amélioré avec l'introduction de Q-SAM2, une méthode de quantification à faible bit qui réduit considérablement les coûts computationnels et de mémoire tout en maintenant une haute fidélité. Cette avancée répond aux défis de déploiement de SAM2 sur des appareils à ressources limitées, le rendant plus accessible pour diverses applications.

The Segment Anything Model 2 (SAM2) has been enhanced with the introduction of Q-SAM2, a low-bit quantization method that significantly reduces computational and memory costs while maintaining high fidelity. This advancement addresses the challenges of deploying SAM2 on resource-constrained devices, making it more accessible for various applications.

Q-SAM2: Accurate Quantization for Segment Anything Model 2

arXiv:2511.18533v1 Announce Type: new 
Abstract: Accurate segmentation of individual teeth from panoramic radiographs remains a challenging task due to anatomical variations, irregular tooth shapes, and overlapping structures. These complexities often limit the performance of conventional deep learning models. To address this, we propose DE-KAN, a novel Dual Encoder Kolmogorov Arnold Network, which enhances feature representation and segmentation precision. The framework employs a ResNet-18 encoder for augmented inputs and a customized CNN encoder for original inputs, enabling the complementary extraction of global and local spatial features. These features are fused through KAN-based bottleneck layers, incorporating nonlinear learnable activation functions derived from the Kolmogorov Arnold representation theorem to improve learning capacity and interpretability. Extensive experiments on two benchmark dental X-ray datasets demonstrate that DE-KAN outperforms state-of-the-art segmentation models, achieving mIoU of 94.5%, Dice coefficient of 97.1%, accuracy of 98.91%, and recall of 97.36%, representing up to +4.7% improvement in Dice compared to existing methods.

تم تقديم إطار عمل جديد يسمى DE-KAN، يستخدم شبكة كولموغوروف أرنولد ذات الترميز المزدوج لتحسين دقة تقسيم الأسنان ثنائية الأبعاد من الأشعة السينية البانورامية. تتناول هذه الطريقة التحديات مثل الاختلافات التشريحية والهياكل المتداخلة التي أعاقت تاريخياً أداء التقسيم. يجمع الإطار بين مشفر ResNet-18 للمدخلات المعززة ومشفر CNN مخصص للمدخلات الأصلية، مما يسمح بتحسين استخراج الميزات.

Se ha introducido un nuevo marco llamado DE-KAN, que utiliza una Red Kolmogorov Arnold de Doble Codificador para mejorar la precisión de la segmentación de dientes en 2D a partir de radiografías panorámicas. Este enfoque aborda desafíos como las variaciones anatómicas y las estructuras superpuestas que históricamente han obstaculizado el rendimiento de la segmentación. El marco combina un codificador ResNet-18 para entradas aumentadas y un codificador CNN personalizado para entradas originales, permitiendo una mejor extracción de características.

Un nouveau cadre nommé DE-KAN a été introduit, utilisant un réseau Kolmogorov Arnold à double encodeur pour améliorer la précision de la segmentation des dents en 2D à partir de radiographies panoramiques. Cette approche répond à des défis tels que les variations anatomiques et les structures qui se chevauchent, qui ont historiquement entravé la performance de la segmentation. Le cadre combine un encodeur ResNet-18 pour les entrées augmentées et un encodeur CNN personnalisé pour les entrées originales, permettant une meilleure extraction des caractéristiques.

A new framework named DE-KAN has been introduced, utilizing a Dual Encoder Kolmogorov Arnold Network to enhance the accuracy of 2D teeth segmentation from panoramic radiographs. This approach addresses challenges such as anatomical variations and overlapping structures that have historically hindered segmentation performance. The framework combines a ResNet-18 encoder for augmented inputs and a customized CNN encoder for original inputs, allowing for improved feature extraction.

DE-KAN: A Kolmogorov Arnold Network with Dual Encoder for accurate 2D Teeth Segmentation

arXiv:2503.22174v3 Announce Type: replace 
Abstract: Intraoperative bleeding in laparoscopic surgery causes rapid obscuration of the operative field to hinder the surgical process and increases the risk of postoperative complications. Intelligent detection of bleeding areas can quantify the blood loss to assist decision-making, while locating bleeding points helps surgeons quickly identify the source of bleeding and achieve hemostasis in time to improve surgical success rates. To fill the benchmark gap, we first construct a real-world laparoscopic surgical bleeding detection dataset, named SurgBlood, comprising 5,330 frames from 95 surgical video clips with bleeding region and point annotations. Accordingly, we develop a dual-task synergistic online detector called BlooDet, enabling simultaneous detection of bleeding regions and points in laparoscopic surgery. The baseline embraces a dual-branch bidirectional guid- ance design based on Segment Anything Model 2. The mask branch detects bleeding regions through adaptive edge and point prompt embeddings, while the point branch leverages mask memory to induce bleeding point memory modeling and captures point motion direction via inter-frame optical flow. By coupled bidirectional guidance, our framework explores spatial-temporal correlations while exploiting memory modeling to infer current bleeding status. Extensive experiments indicate that our method outperforms 13 counterparts in bleeding detection.

تم تطوير كاشف عبر الإنترنت ذو مهام مزدوجة يسمى BlooDet لتحسين الكشف عن مناطق ونقاط النزيف في مقاطع الفيديو الجراحية بالمنظار. تعالج هذه الابتكار القضية الحرجة للنزيف أثناء العملية، الذي يمكن أن يعيق مجال الجراحة ويعقد الإجراءات، مما يؤدي إلى زيادة محتملة في المضاعفات بعد الجراحة. كما تقدم الدراسة مجموعة بيانات SurgBlood، التي تتضمن 5,330 إطارًا من 95 مقطع فيديو جراحي تم وضع علامات عليها للكشف عن النزيف.

Se ha desarrollado un nuevo detector en línea sinérgico de doble tarea llamado BlooDet para mejorar la detección de regiones y puntos de sangrado en videos quirúrgicos laparoscópicos. Esta innovación aborda el problema crítico del sangrado intraoperatorio, que puede oscurecer el campo quirúrgico y complicar los procedimientos, lo que potencialmente lleva a un aumento de las complicaciones postoperatorias. La investigación también presenta el conjunto de datos SurgBlood, que incluye 5,330 fotogramas de 95 clips de video quirúrgicos anotados para la detección de sangrado.

Un nouveau détecteur en ligne synergique à double tâche, nommé BlooDet, a été développé pour améliorer la détection des régions et des points de saignement dans les vidéos chirurgicales laparoscopiques. Cette innovation répond à la problématique critique du saignement intraopératoire, qui peut obscurcir le champ opératoire et compliquer les procédures, entraînant potentiellement des complications postopératoires accrues. La recherche introduit également le jeu de données SurgBlood, qui comprend 5 330 images de 95 clips vidéo chirurgicaux annotés pour la détection des saignements.

A new dual-task synergistic online detector named BlooDet has been developed to enhance the detection of bleeding regions and points in laparoscopic surgical videos. This innovation addresses the critical issue of intraoperative bleeding, which can obscure the surgical field and complicate procedures, potentially leading to increased postoperative complications. The research also introduces the SurgBlood dataset, which includes 5,330 frames from 95 surgical video clips annotated for bleeding detection.

Q-SAM2: Accurate Quantization for Segment Anything Model 2

Was this article worth reading? Share it

LangWatch

AQ

Clarity.Tube