EvaSurf: Efficient View-Aware Implicit Textured Surface Reconstruction.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computer Graphics

Background:

• Traditional 3D reconstruction methods struggle with real-time performance and high-fidelity output.
• Differentiable rendering techniques like Neural Radiance Fields (NeRF) offer high-fidelity but are computationally expensive.
• Existing methods are often impractical for daily applications due to rendering runtime limitations.

Purpose of the Study:

• To develop an efficient 3D reconstruction method suitable for mobile devices.
• To achieve high-fidelity mesh reconstruction with accurate, view-dependent textures.
• To enable real-time rendering performance on common mobile hardware.

Main Methods:

• Introduced EvaSurf, an Efficient View-Aware implicit textured Surface reconstruction method.
• Employed an efficient surface-based model with multi-view supervision for accurate mesh generation.
• Learned an implicit texture with view-aware encoding for high-fidelity, view-dependent rendering.
• Utilized a lightweight neural shader for reduced computational cost and real-time performance.

Main Results:

• EvaSurf reconstructs high-quality appearance and accurate meshes on both synthetic and real-world datasets.
• The method achieves real-time performance on mobile devices at over 40 Frames Per Second (FPS).
• Training is efficient, requiring only 1-2 hours on a single GPU, with a small final package size (40-50 MB).

Conclusions:

• EvaSurf offers a practical solution for real-time 3D reconstruction on mobile devices.
• The method successfully balances high-fidelity appearance and accurate geometry with computational efficiency.
• This approach significantly advances the applicability of 3D reconstruction in everyday mobile scenarios.