GarmentRec: Towards Individual Garment Reconstruction From a Monocular Human Image

Area of Science:

• Computer Vision
• 3D Graphics
• Geometric Modeling

Background:

• Implicit methods struggle with garment mesh reconstruction from single images, often yielding incomplete or distorted results.
• Explicit parametric models offer robust shape constraints but lack detail and variation.
• Existing methods fail to capture intricate garment features like wrinkles and pockets effectively.

Purpose of the Study:

• To develop a novel explicit garment template capable of reconstructing detailed 3D garment models from monocular images.
• To address the limitations of existing methods in handling both closed and open garment topologies and capturing fine geometric details.
• To enable accurate and robust garment reconstruction for applications like human digitization and virtual try-on.

Main Methods:

• Introduction of a new explicit garment template supporting both closed and open topologies.
• Proposal of a detailed garment reconstruction method utilizing the new template from monocular views.
• Prediction of displacement maps on the parameterization domain and elaboration to 3D surfaces via UV coordinates for detail recovery.

Main Results:

• The proposed method successfully reconstructs individual garment meshes from monocular images.
• Realistic details such as wrinkles and pockets are faithfully recovered in an explicit manner.
• Demonstrated accuracy and robustness across extensive experiments for garment shape recovery.

Conclusions:

• The novel explicit garment template and reconstruction method provide a significant advancement in 3D garment modeling.
• The approach effectively overcomes previous limitations, enabling detailed and accurate reconstruction of various garment types.
• The method shows promise for practical applications in human digitization and virtual try-on scenarios.