Texture-Consistent 3D Scene Style Transfer via Transformer-Guided Neural Radiance Fields

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• Neural Radiance Fields (NeRFs) show promise for 3D style transfer.
• Existing NeRF-based methods struggle with preserving scene textures and cross-view consistency.

Purpose of the Study:

• To develop a novel transformer-guided approach for 3D scene style transfer.
• To address limitations in texture preservation and cross-view consistency in current methods.

Main Methods:

• A transformer-based network generates 2D stylized images for supervision.
• A latent style vector and style network enable fine-grained style control in 3D.
• A merge network integrates style features with scene geometry for rendering.
• A texture consistency loss is introduced to maintain fidelity across views.

Main Results:

• The proposed method achieves superior visual perception and image quality.
• Experimental results demonstrate enhanced multi-view consistency.
• The approach outperforms state-of-the-art methods in quantitative and qualitative evaluations.

Conclusions:

• The transformer-guided approach effectively addresses challenges in NeRF-based 3D style transfer.
• The method successfully preserves scene textures and maintains strong cross-view consistency.
• This work advances the capabilities of 3D scene stylization.