计算机图形学

基础概念

对比学习（Contrastive Learning）

Computer Graphics

Convex Hall

1. Sort

2. Color modulation

3. Postman problem

Voronoi

MAT (Medial Axis Transformation)

中轴变换，用于简化三维物体的表示

缺点：对于物体表面的噪声敏感

Voronoi Diagram (VD)

DT的对偶表示

Delaunay

Delaunay Trangulation (DT)

将一个多边形进行三角化

希望所有的三角形越接近正三角形越好

Mesh

Geometry Processing

• Downsampling (Compression)
• Upsampling
  • Chaikin's algorithm
  • B-spline

Morphing (变换)

Mesh Simplification

Interpolation (插值)

Affine Transformation (仿射变换)

• translation
• scale (缩放)
• rotation (旋转)
• shear (错切)
• flip (翻转)

Mesh Reconstruction and Smoothing

• PointNet

  提取三维点云中的特征（输入是点的三维坐标）

• VexelNet

  先将点云转换成体素网格，然后提取体素网格的特征

3D Detection from Point Clouds

• Multi-class Object Detection
• Network for Crowd Perception
• LiDAR-based methods
• Fusion-based methods
  • Transfusion (Transformer-based)

Human Motion Capture

SMPL Parameters (参数化表示)

Sensors-Data

• Wearable

  • Infrared Cameras(红外相机) + Markers -> Keypoints
  • IMUs (Inertial Measurement Units) -> Rot and Acc

• Vision

  • Camera -> Image
  • LiDAR -> Point Cloud

Marker-based

• SOMA (Solving Optical Marker-Based MoCap Automatically)

IMU-based

• TransPose (sparse IMUs)

Camera-based

• GVHMR

LiDAR-based

• LiveHPS
• LiveHPS++