v2.0.0 - Segmentation

This release adds an automatic semantic labelling stage to the pipeline. The Docker image has also been roughly halved in size.
The pipeline now also writes scene_semantic.ply and labels.json. Pass --no-semantic for the previous reconstruction-only behaviour.

v1.0.0 — Initial release

First working release of Video to 3D Reconstruction

What's included

• Feed-forward 3D reconstruction from a phone video using AMB3R (CVPR 2026)
• Interactive point cloud viewer (Open3D)
• Outputs: .ply point cloud + Nerfstudio-compatible camera poses
• Fully containerised — runs inside Docker, no manual environment setup

Requirements

• NVIDIA GPU (8GB+ VRAM)
• Docker + NVIDIA Container Toolkit

Coming in v2.0.0

• Semantic labels