View-Adaptive Renderer for View-Consistent 2D-to-3D Generation
2026-08-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionGraphics
AI summaryⓘ
The authors address the challenge of creating 3D shapes from just one image, which is difficult because the views generated from a single image often don't match perfectly. They developed a new method where special neural renderers fix errors specific to each viewpoint while sharing overall shape information to keep things consistent. They also use a self-attention module to combine different views smoothly without needing complex or slow techniques. Their experiments show that this method improves the accuracy of 3D reconstruction while being more efficient and practical.
3D reconstructionsingle image 3D generationNeural Radiance Fields (NeRF)viewpoint adaptationneural renderingself-attention fusionphotometric rendering lossmulti-view synthesisgeometric consistency
Authors
U-Chae Jun, Jaeeun Ko, Jiwoo Kang
Abstract
Reconstructing 3D shapes from a single image remains a fundamental yet challenging problem in computer vision. Traditional monocular 3D generation pipelines typically synthesize multiple views from a single input image before applying Neural Radiance Field (NeRF)-based reconstruction. However, inherent projective ambiguities often produce visual discontinuities across generated viewpoints, leading to inaccuracies in reconstructed 3D models. Current solutions either incur significant additional computational burdens or fail to adequately resolve practical inconsistencies between synthesized views. To address these limitations, we propose a novel viewpoint-adaptive neural rendering framework that enables robust 3D reconstruction even when given partially inconsistent multi-view inputs. Our approach introduces view-adaptive neural renderers that independently correct viewpoint-dependent errors while simultaneously sharing a global feature backbone to preserve structural coherence. Furthermore, we propose a self-attention fusion module that adaptively integrates multi-view information, ensuring geometric consistency without relying heavily on indirect regularizations or computationally intensive methods. Through extensive experiments, we demonstrate that our method consistently improves 3D reconstruction fidelity. Importantly, our approach achieves near state-of-the-art performance without diffusion-based SDS supervision, relying primarily on photometric rendering loss with lightweight attention regularizers. This balance between accuracy and efficiency makes the proposed framework highly practical for real-world applications.