Photorealistic style transfer in neural radiance fields (NeRF) aims to modify the color characteristics of a 3D scene without altering its underlying geometry. Although recent approaches have achieved promising results, they often suffer from limited style diversity, focusing primarily on global color shifts. In contrast, artistic style transfer methods offer richer stylization but usually distort scene geometry, thereby reducing realism.
In this work, we present Intrinsic-guided Photorealistic Style Transfer (IPRF), a novel framework that leverages intrinsic image decomposition to decouple a scene into albedo and shading components. By introducing tailored loss functions in both domains, IPRF aligns the texture and color of the content scene to those of a style image while faithfully preserving geometric structure and lighting.
Furthermore, we propose Tuning-assisted Style Interpolation (TSI), a real-time technique for exploring the trade-off between photorealism and artistic expression through a weighted combination of albedo-oriented and shading-oriented radiance fields. Experimental results demonstrate that IPRF achieves a superior balance between naturalism and artistic expression compared to state-of-the-art methods, offering a versatile solution for 3D content creation in various fields, including digital art, virtual reality, and game design.
IPRF extracts albedo and shading from rendered and style images via neural intrinsic image decomposition and integrates them into the style‑transfer loss. Albedo encodes color that is invariant to lighting, whereas shading captures shadows that depend on illumination and viewpoint. Using a pretrained network, IPRF predicts these components and quantitatively compares their characteristics between the style and content images with VGG features. The overall objective includes albedo‑ and shading‑matching terms together with a total variation (TV) regularizer to maintain geometric and textural consistency during transfer. As a result, the method achieves photorealistic style transfer that preserves the original scene’s structure and texture while transferring the style image’s colors and the content image’s shadow characteristics.
TSI generates diverse styles by adjusting the contributions of two intrinsic image components—albedo and shading—without any additional training. It leverages two optimized radiance fields: one that models style albedo only and another that models content shading only. By linearly combining these fields with a user-controlled weight, TSI provides continuous style control within a 3D scene, enabling real-time exploration without costly parameter re-optimization. This approach lets users intuitively tune styles and quickly preview the results.
Coming Soon!