Anisotropic Spherical Gaussians Lighting Priors for Indoor Environment Map Estimation

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• High Dynamic Range (HDR) environment lighting is crucial for realistic augmented reality (AR) and visual editing.
• Acquiring accurate HDR maps is challenging, requiring specialized equipment and post-processing.
• Current deep learning methods struggle with complex indoor lighting variations.

Purpose of the Study:

• To develop a novel method for estimating indoor HDR environment maps from single standard images.
• To leverage Anisotropic Spherical Gaussians (ASG) for modeling complex lighting distributions.
• To improve the realism of relighting and scene composition in AR and visual editing applications.

Main Methods:

• Utilized Anisotropic Spherical Gaussians (ASG) to represent intricate lighting distributions as priors.
• Developed a transformer-based network with a two-stage training scheme for ASG parameter prediction.
• Introduced a generative projector for synthesizing high-frequency textures in environment maps.
• Employed a parameter-efficient adaptation method to transfer knowledge from Spherical Gaussian (SG) to ASG.

Main Results:

• The proposed method effectively estimates ASG parameters for indoor lighting.
• The generative projector successfully synthesizes environment maps with high-frequency textures.
• The ASG-based approach captures fine-grained anisotropic lighting characteristics more effectively than traditional SG.
• Experimental results show improved precision in lighting conditions and environment textures.

Conclusions:

• The novel method accurately estimates indoor HDR environment maps from single images using ASG.
• The approach enhances the realism of lighting effects in AR and visual editing.
• This work offers a more efficient and effective solution for capturing complex indoor lighting compared to existing methods.