A public coursework archive for a 3D Gaussian Splatting team project completed in the “3D From NeRF to World Models” summer course.

https://github.com/ANHOYA/3DGS_HYU_AI_bootcamp

Overview

• Course: AI-based 3D Autonomous Driving — 3D From NeRF to World Models
• Semester: Summer 2026
• Department: Electrical Engineering, Hanyang University
• Project type: Team project
• Focus: 3D reconstruction, Structure from Motion, Multi-View Stereo, NeRF, 3D Gaussian Splatting, and World Models
• Main tools: COLMAP, 3D Gaussian Splatting, CUDA, Conda

Slides

1조_팀프로젝트_발표자료.pdf

Video

https://youtu.be/dR50moSzY7M

Project — 3D Gaussian Splatting Reconstruction

Our team built an end-to-end reconstruction pipeline from a short 720p video captured with the Blackmagic Camera app on an iPhone/iPad. We extracted frames, estimated camera poses and sparse geometry with COLMAP, undistorted the images, and trained 3D Gaussian Splatting models for several indoor scenes.

Pipeline

Captured video
→ frame extraction
→ COLMAP camera-pose estimation and sparse reconstruction
→ COLMAP undistortion
→ 3D Gaussian Splatting training
→ point-cloud and Gaussian reconstruction review