ePBR Material Enables Real-Time High-Fidelity Image Synthesis

Realistic indoor and outdoor image synthesis remains a central challenge in computer vision and graphics. Learning-based methods are user-friendly but often lack physical consistency. On the other hand, traditional physics-based rendering (PBR) techniques can produce highly realistic images but come with substantial computational costs. Intrinsic image representation offers a promising compromise by decomposing images into their fundamental components, such as geometry, material, and lighting, allowing for controlled synthesis. However, existing PBR materials struggle with complex surface models, particularly those involving high specular reflections and transparent surfaces. In this research, we extend intrinsic image representation to include reflection and transmission properties, enabling the synthesis of materials like glass and windows. We introduce a framework called extended PBR (ePBR) materials, which provides deterministic and interpretable image synthesis. This approach allows for effective editing and precise control over the final rendered images. The ePBR framework addresses a significant limitation of traditional PBR by incorporating detailed reflection and transmission attributes. These attributes are crucial for accurately representing materials with transparent or highly reflective surfaces. By breaking down the image synthesis process into these specific channels, we achieve greater flexibility and control, making it easier to edit and manipulate materials in the rendered environment. To demonstrate the effectiveness of our method, we conducted several experiments. The results show that ePBR materials can handle a wide range of surface types, including glass, water, and metallic objects, with enhanced realism and physical accuracy. This not only improves the visual fidelity of the synthesized images but also ensures that they align with real-world physics, making them more useful for applications such as virtual reality, augmented reality, and architectural visualization. In conclusion, our work on ePBR materials represents a significant step forward in the field of image synthesis. By integrating reflection and transmission properties into the intrinsic image representation, we offer a solution that balances the ease of use of learning-based methods with the physical accuracy of PBR techniques. This advancement promises to broaden the scope of realistic image synthesis, opening up new possibilities for various industries and research areas.