מושגי ליבה
PoNQ introduces a novel learnable mesh representation using Quadric Error Metrics (QEM) for efficient training and sharp reconstructions.
תקציר
Polygon meshes are standard in geometry processing but have limitations for learning-based applications due to their irregular nature.
PoNQ utilizes local 3D sample points, normals, and QEM to create a global mesh efficiently.
Guarantees topological and geometrical properties, ensuring no self-intersections and always forming the boundary of a volume.
Outperforms state-of-the-art techniques in mesh prediction from SDF grids.
Detailed explanation of the method, including optimization, learning tasks, and mesh extraction process.
סטטיסטיקה
PoNQは、Quadric Error Metric(QEM)を使用して効率的なトレーニングと鮮明な再構築を行うための新しい学習可能なメッシュ表現を導入します。