Pytorch3D

Structures

• Meshes: vertices + faces + textures
• PointCloud: points + rgb

Cameras

• Camera Coordinate System
• look_at_view_transform: distance + elevation + azimuth（相机与物体的距离+高度角+方位角，默认为角度制，degrees=False 置为弧度制）→ R + T
• FoVPerspectiveCameras: Rotation（旋转矩阵 R）+ Translation（平移变换 T）+ Field-of-View（视野 fov）

Renderer

• Renderer: structures + cameras + lightings → output: $B\times H\times W\times 4$ ( $RGB \in [0,1]$ + 轮廓信息)
• MeshRenderer: rasterizer + shader

Representations

• Depth Maps & Normal Maps (2.5D)
• Surface Representations
• Volume Representations
• Summary: 不同表示方法的抉择

Representations Conversions

• 重点：Differentiable Transformations

  

• Conversions between volume representations

• Conversions between surface representations