arXiv:2511.14310v1 Announce Type: new 
Abstract: Multi-source stationary computed tomography (CT) has recently attracted attention for its ability to achieve rapid image reconstruction, making it suitable for time-sensitive clinical and industrial applications. However, practical systems are often constrained by ultra-sparse-view sampling, which significantly degrades reconstruction quality. Traditional methods struggle under ultra-sparse-view settings, where interpolation becomes inaccurate and the resulting reconstructions are unsatisfactory. To address this challenge, this study proposes Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework tailored for multi-source stationary CT under ultra-sparse-view conditions. Diff-NAF combines a Neural Attenuation Field representation with a dual-branch conditional diffusion model. The process begins by training an initial NAF using ultra-sparse-view projections. New projections are then generated through an Angle-Prior Guided Projection Synthesis strategy that exploits inter view priors, and are subsequently refined by a Diffusion-driven Reuse Projection Refinement Module. The refined projections are incorporated as pseudo-labels into the training set for the next iteration. Through iterative refinement, Diff-NAF progressively enhances projection completeness and reconstruction fidelity under ultra-sparse-view conditions, ultimately yielding high-quality CT reconstructions. Experimental results on multiple simulated 3D CT volumes and real projection data demonstrate that Diff-NAF achieves the best performance under ultra-sparse-view conditions.

لقد جذبت التصوير المقطعي المحوسب الثابت متعدد المصادر مؤخرًا الانتباه لقدرتها على تحقيق إعادة بناء الصور بسرعة، مما يجعلها مناسبة للتطبيقات السريرية والصناعية الحساسة للوقت. ومع ذلك، غالبًا ما يؤدي أخذ العينات ذات العرض النادر للغاية إلى تدهور كبير في جودة إعادة البناء. تكافح الطرق التقليدية في هذه الإعدادات، حيث تصبح الاستيفاءات غير دقيقة وتكون إعادة البناء الناتجة غير مرضية. لمعالجة هذا التحدي، تقترح هذه الدراسة حقول التخفيف العصبي المكررة بواسطة الانتشار (Diff-NAF)، وهو إطار تكراري مصمم للتصوير المقطعي الثابت متعدد المصادر في ظل ظروف العرض النادر للغاية.

La tomografía computarizada (CT) estacionaria de múltiples fuentes ha atraído recientemente la atención por su capacidad para lograr reconstrucciones de imágenes rápidas, adecuadas para aplicaciones clínicas e industriales sensibles al tiempo. Sin embargo, el muestreo ultraesparcido a menudo conduce a una degradación significativa en la calidad de la reconstrucción. Los métodos tradicionales luchan en estas configuraciones, lo que resulta en interpolaciones inexactas y reconstrucciones insatisfactorias. Para abordar este desafío, el estudio propone Campos de Atenuación Neuronal Refinados por D…

La tomographie par ordinateur (CT) stationnaire multi-sources a récemment suscité de l'intérêt en raison de sa capacité à réaliser des reconstructions d'images rapides, adaptées aux applications cliniques et industrielles sensibles au temps. Cependant, l'échantillonnage ultra-sparse dégrade souvent la qualité de reconstruction. Les méthodes traditionnelles peinent dans ces conditions, entraînant des interpolations inexactes et des images insatisfaisantes. Pour surmonter cette limitation, l'étude propose des Champs d'Atténuation Neuronale Raffinés par Diffusion (Diff-NAF), un cadre itératif con…

Multi-source stationary computed tomography (CT) has gained attention for its rapid image reconstruction capabilities, particularly in time-sensitive clinical and industrial applications. However, ultra-sparse-view sampling often leads to significant degradation in reconstruction quality. Traditional methods struggle in these settings, resulting in inaccurate interpolations and unsatisfactory images. To overcome this limitation, the study introduces Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework designed for multi-source stationary CT under ultra-sparse-view con…

Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

arXiv:2511.18452v1 Announce Type: new 
Abstract: Vision Foundation Models (VFMs) extract spatially downsampled representations, posing challenges for pixel-level tasks. Existing upsampling approaches face a fundamental trade-off: classical filters are fast and broadly applicable but rely on fixed forms, while modern upsamplers achieve superior accuracy through learnable, VFM-specific forms at the cost of retraining for each VFM. We introduce Neighborhood Attention Filtering (NAF), which bridges this gap by learning adaptive spatial-and-content weights through Cross-Scale Neighborhood Attention and Rotary Position Embeddings (RoPE), guided solely by the high-resolution input image. NAF operates zero-shot: it upsamples features from any VFM without retraining, making it the first VFM-agnostic architecture to outperform VFM-specific upsamplers and achieve state-of-the-art performance across multiple downstream tasks. It maintains high efficiency, scaling to 2K feature maps and reconstructing intermediate-resolution maps at 18 FPS. Beyond feature upsampling, NAF demonstrates strong performance on image restoration, highlighting its versatility. Code and checkpoints are available at https://github.com/valeoai/NAF.

يمثل إدخال فلترة الانتباه الجوار (NAF) تقدمًا كبيرًا في مجال نماذج الأساس البصرية (VFMs)، مما يسمح بزيادة الميزات بدون إعادة تدريب. تستخدم هذه الطريقة المبتكرة الانتباه الجوار عبر المقاييس وموضع التدوير لتعلم أوزان مكانية ومحتوى بشكل تكيفي من الصور عالية الدقة، متفوقة على نماذج الزيادة الخاصة بـ VFM عبر مهام متعددة.

La introducción del Filtrado de Atención de Vecindario (NAF) representa un avance significativo en el campo de los Modelos de Fundación Visual (VFM), permitiendo el aumento de características en cero disparos sin necesidad de reentrenamiento. Este método innovador utiliza la Atención de Vecindario a Escalas Cruzadas y los Embeddings de Posición Rotativa para aprender de manera adaptativa los pesos espaciales y de contenido a partir de imágenes de alta resolución, superando a los aumentadores específicos de VFM en diversas tareas.

L'introduction du filtrage par attention de voisinage (NAF) représente une avancée significative dans le domaine des modèles de fondation visuelle (VFM), permettant un suréchantillonnage de caractéristiques en zéro-shot sans besoin de réentraînement. Cette méthode innovante utilise l'attention de voisinage à échelles croisées et les embeddings de position rotatifs pour apprendre de manière adaptative des poids spatiaux et de contenu à partir d'images haute résolution, surpassant les suréchantillonneurs spécifiques aux VFM dans diverses tâches.

The introduction of Neighborhood Attention Filtering (NAF) represents a significant advancement in the field of Vision Foundation Models (VFMs), allowing for zero-shot feature upsampling without the need for retraining. This innovative method utilizes Cross-Scale Neighborhood Attention and Rotary Position Embeddings to adaptively learn spatial and content weights from high-resolution images, outperforming existing VFM-specific upsamplers across various tasks.

NAF: Zero-Shot Feature Upsampling via Neighborhood Attention Filtering

arXiv:2511.18806v1 Announce Type: new 
Abstract: X-ray imaging, based on penetration, enables detailed visualization of internal structures. Building on this capability, existing implicit 3D reconstruction methods have adapted the NeRF model and its variants for internal CT reconstruction. However, these approaches often neglect the significance of objects' anatomical priors for implicit learning, limiting both reconstruction precision and learning efficiency, particularly in ultra-sparse view scenarios. To address these challenges, we propose a novel 3D CT reconstruction framework that employs a 'target prior' derived from the object's projection data to enhance implicit learning. Our approach integrates positional and structural encoding to facilitate voxel-wise implicit reconstruction, utilizing the target prior to guide voxel sampling and enrich structural encoding. This dual strategy significantly boosts both learning efficiency and reconstruction quality. Additionally, we introduce a CUDA-based algorithm for rapid estimation of high-quality 3D target priors from sparse-view projections. Experiments utilizing projection data from a complex abdominal dataset demonstrate that the proposed model substantially enhances learning efficiency, outperforming the current leading model, NAF, by a factor of ten. In terms of reconstruction quality, it also exceeds the most accurate model, NeRP, achieving PSNR improvements of 3.57 dB, 5.42 dB, and 5.70 dB with 10, 20, and 30 projections, respectively. The code is available at https://github.com/qlcao171/TPG-INR.

تم تقديم إطار جديد يسمى TPG-INR لإعادة بناء التصوير المقطعي ثلاثي الأبعاد، مما يعزز التعلم الضمني من خلال استخدام 'مبدأ مستهدف' مستمد من بيانات الإسقاط. تهدف هذه الطريقة إلى تحسين دقة وكفاءة إعادة البناء، لا سيما في السيناريوهات ذات العرض النادر للغاية، من خلال دمج الترميز المكاني والهيكلي لإعادة البناء على مستوى الفوكسل.

Se ha introducido un nuevo marco llamado TPG-INR para la reconstrucción de CT 3D, que mejora el aprendizaje implícito al utilizar un 'prior objetivo' derivado de los datos de proyección. Este método busca mejorar la precisión y eficiencia de la reconstrucción, especialmente en escenarios de vista ultra-esparcida, integrando codificación posicional y estructural para la reconstrucción voxel por voxel.

Un nouveau cadre appelé TPG-INR a été introduit pour la reconstruction CT 3D, améliorant l'apprentissage implicite en utilisant un 'prior cible' dérivé des données de projection. Cette méthode vise à améliorer la précision et l'efficacité de la reconstruction, en particulier dans des scénarios de vue ultra-sparse, en intégrant un encodage positionnel et structurel pour la reconstruction voxel par voxel.

A novel framework called TPG-INR has been introduced for 3D CT reconstruction, enhancing implicit learning by utilizing a 'target prior' derived from projection data. This method aims to improve reconstruction precision and efficiency, particularly in ultra-sparse view scenarios, by integrating positional and structural encoding for voxel-wise reconstruction.

Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

Was this article worth reading? Share it

GPTHumanizer

Zemith-3bda3b

AINF Membership