3D Computer Vision Study Plan
A comprehensive 13-week curriculum covering 3D computer vision from geometric foundations to neural rendering. Camera geometry, stereo vision, point clouds, NeRF, 3D Gaussian Splatting, and generative 3D.
Recommended Study Path
Foundations
Weeks 1-2
- Ch 1: 3D Basics & Coordinate Systems
- Ch 2: Camera Models & Calibration
Representations, transforms, camera geometry
Geometry & Stereo
Weeks 3-4
- Ch 3: Epipolar & Stereo
- Ch 4: Multi-View Geometry
Fundamental matrix, SfM, bundle adjustment
Reconstruction
Weeks 5-7
- Ch 5: Depth Estimation
- Ch 6: 3D Reconstruction
- Ch 7: Point Clouds
Dense reconstruction, mesh generation, PointNet
3D Understanding
Weeks 8-10
- Ch 8: 3D Detection
- Ch 9: Scene Understanding
- Ch 10: SLAM
Detection, segmentation, real-time mapping
Neural 3D
Weeks 11-13
- Ch 11: NeRF
- Ch 12: 3D Gaussian Splatting
- Ch 13: Generative 3D
Neural rendering, novel view synthesis, text-to-3D
All Chapters
Introduction to 3D Computer Vision
3D representations, coordinate systems, rigid transformations, and the foundations of spatial computing.
Camera Models & Calibration
Pinhole camera model, lens distortion, intrinsic/extrinsic parameters, and calibration techniques.
Epipolar Geometry & Stereo Vision
Fundamental and essential matrices, stereo rectification, matching algorithms, and disparity estimation.
Multi-View Geometry
Structure from Motion, bundle adjustment, COLMAP pipelines, and multi-view reconstruction.
Depth Estimation
Monocular depth networks, multi-view stereo, active sensing with LiDAR and structured light.
3D Reconstruction Pipelines
Photogrammetry, mesh generation, Poisson reconstruction, signed distance functions, and texture mapping.
Point Cloud Processing
PointNet architectures, 3D convolutions, voxelization, segmentation, and registration with ICP.
3D Object Detection
VoxelNet, PointPillars, CenterPoint, BEV detection, sensor fusion, and evaluation.
3D Scene Understanding
Semantic and instance segmentation in 3D, scene graphs, occupancy networks, and indoor reconstruction.
Visual SLAM & Localization
Feature-based and direct SLAM, visual-inertial odometry, deep SLAM, relocalization, and loop closure.
Neural Radiance Fields (NeRF)
Volume rendering, positional encoding, Instant-NGP, Mip-NeRF, and dynamic scene extensions.
3D Gaussian Splatting
Gaussian representation, differentiable rasterization, optimization, densification, and dynamic 4D extensions.
Generative 3D Models
Text-to-3D with DreamFusion, image-to-3D reconstruction, multi-view diffusion, and 3D-aware generation.
Curriculum inspired by state-of-the-art 3D vision research, designed to take you from geometric foundations to neural rendering.