Interactive High-Relief Reconstruction for Organic and Double-Sided Objects from a Photo

Area of Science:

• Computer Vision
• Computer Graphics
• 3D Reconstruction

Background:

• Reconstructing high-relief 3D geometry from single images is challenging, especially for non-rigid, organic objects and double-sided structures.
• Existing methods struggle with complex shapes, occlusions, and simultaneous front/back view reconstruction.

Purpose of the Study:

• To introduce an interactive, user-driven computational pipeline for reconstructing high-relief 3D geometry from a single photograph.
• To address the challenges of non-rigid object shapes and double-sided structures in 3D reconstruction.

Main Methods:

• A three-stage pipeline involving 2.5D model construction (segmentation, layering, occlusion handling), user-guided inflation (slope/curvature cues, optimization), and final stitching/optimization.
• Interactive editing with real-time preview of inflated geometry.
• Constrained optimization model guided by user-provided cues.

Main Results:

• Successful reconstruction of high-relief 3D geometry from single images, even with large viewing angles.
• Ability to handle complex organic objects, multiple occlusions, and varying shape profiles.
• Reconstruction of objects with double-sided structures, a significant advancement over prior work.

Conclusions:

• The proposed method enables high-quality 3D reconstruction of challenging objects from single images.
• Demonstrated applicability across diverse subjects including humans, animals, and flowers.
• Offers a robust solution for generating detailed 3D models with significant depth and complexity.