arXiv:2511.18672v1 Announce Type: new 
Abstract: Novel View Synthesis (NVS) is the task of generating new images of a scene from viewpoints that were not part of the original input. Diffusion-based NVS can generate high-quality, temporally consistent images, however, remains computationally prohibitive. Conversely, regression-based NVS offers suboptimal generation quality despite requiring significantly lower compute; leaving the design objective of a high-quality, inference-efficient NVS framework an open challenge. To close this critical gap, we present Sphinx, a training-free hybrid inference framework that achieves diffusion-level fidelity at a significantly lower compute. Sphinx proposes to use regression-based fast initialization to guide and reduce the denoising workload for the diffusion model. Additionally, it integrates selective refinement with adaptive noise scheduling, allowing more compute to uncertain regions and frames. This enables Sphinx to provide flexible navigation of the performance-quality trade-off, allowing adaptation to latency and fidelity requirements for dynamically changing inference scenarios. Our evaluation shows that Sphinx achieves an average 1.8x speedup over diffusion model inference with negligible perceptual degradation of less than 5%, establishing a new Pareto frontier between quality and latency in NVS serving.

Sphinx هو إطار استدلال هجين تم تقديمه حديثًا مصمم لتحسين توليد المناظر الجديدة (NVS) من خلال تحقيق دقة صورة بمستوى انتشار مع تقليل كبير في المتطلبات الحاسوبية. يستخدم هذا الإطار بدءًا سريعًا قائمًا على الانحدار لتوجيه عملية إزالة الضوضاء لنموذج الانتشار، إلى جانب تحسين انتقائي وجدولة ضوضاء تكيفية لتحسين الأداء في المناطق غير المؤكدة.

Sphinx es un nuevo marco de inferencia híbrido diseñado para mejorar la Síntesis de Nuevas Vistas (NVS) al lograr una fidelidad de imagen a nivel de difusión mientras reduce significativamente las demandas computacionales. Este marco emplea una inicialización rápida basada en regresión para guiar el proceso de desruido del modelo de difusión, junto con un refinamiento selectivo y una programación adaptativa de ruido para optimizar el rendimiento en regiones inciertas.

Sphinx est un nouveau cadre d'inférence hybride conçu pour améliorer la synthèse de nouvelles vues (NVS) en atteignant une fidélité d'image de niveau diffusion tout en réduisant considérablement les exigences computationnelles. Ce cadre utilise une initialisation rapide basée sur la régression pour guider le processus de débruitage du modèle de diffusion, associé à un raffinement sélectif et à une planification adaptative du bruit pour optimiser les performances dans les régions incertaines.

Sphinx is a newly introduced hybrid inference framework designed to enhance Novel View Synthesis (NVS) by achieving diffusion-level image fidelity while significantly reducing computational demands. This framework employs regression-based fast initialization to guide the denoising process of the diffusion model, coupled with selective refinement and adaptive noise scheduling to optimize performance in uncertain regions.

Sphinx: Efficiently Serving Novel View Synthesis using Regression-Guided Selective Refinement

Was this article worth reading? Share it

Republiclabs.ai

GPTHumanizer

AIPortalX