RiGS — Repository reference¶

Development line: project:rigs · thread repository-reference
Last event: 2026-07-25 · 1 dated since 2026-07-25 · Researched: 2026-09-05 · confidence: high

What it is¶

RiGS provides the codebase for Rigid-aware 4D Gaussian Splatting: - Primitives: separates scenes into static, rigid, and transient Gaussian components. - Data preparation: generates inputs through ViPE and TAPIR/BootsTAPIR. - Workflow: trains, evaluates, renders, and displays scenes in a viewer.

Requires Python 3.10 and a CUDA 12.x compatible stack. Research code without release packages.

Development line¶

• 2026-07-25 — RiGS project repository was referenced. On 2026-07-25, the project line linked to the RiGS GitHub repository. The reference documents public availability without claiming features, release status, or repository contents.

What changed¶

2026-07-25 — The RiGS implementation repository became available. Source materials specify that the method operates on a single monocular video and uses static, rigid, and transient components across distinct motion timescales.

2026-05-22 — Earlier date: the authors published the RiGS paper on arXiv (v1), describing object-wise dynamic masks and scene-flow supervision.

How to use this¶

From 2026-07-25, use the linked RiGS GitHub repository as the dated source reference for this project line, while verifying its current contents separately.

1. Clone the repository with its ViPE submodule, create a Python 3.10 conda environment, install requirements and the local ViPE package, then download the TAPIR/BootsTAPIR checkpoint. — https://github.com/ladvu/RiGS
2. Prepare video and frames, extract depth, poses, intrinsics, optical flow, flow consistency, and static masks through ViPE, then build TAPIR tracks. — https://github.com/ladvu/RiGS
3. Run src/main.py with data paths and an experiment name; use --eval_step, --render_video, or run_viewer.py to inspect checkpoint results. — https://github.com/ladvu/RiGS

Best practices¶

• Clone with --recursive, or initialize submodules manually afterward; monocular preprocessing will not run without them. — https://github.com/ladvu/RiGS
• Match PyTorch, CUDA toolkit, and driver versions against pinned requirements before training, because the repository targets CUDA 12.x. — https://github.com/ladvu/RiGS
• Inspect preprocessing visualizations and generate TAPIR tracks before reconstruction: the pipeline requires both ViPE outputs and 2D tracks. — https://github.com/ladvu/RiGS

Superseded by this¶

• Nothing marked obsolete yet.

Still unknown¶

• GitHub does not publish releases, so the repository does not provide a versioned release history or a packaged-installation path.
• The available first-party pages do not expose a dated repository-creation or commit record that independently ties the code publication specifically to 2026-07-25; that date remains the dated event supplied for the repository link.

Sources¶

Agent brief¶

• Subject: project:rigs, thread repository-reference, 1 dated events 2026-07-25 → 2026-07-25.
• Practical note: From 2026-07-25, practitioners should use the linked RiGS GitHub repository as the dated source reference for this project line, while verifying its current contents separately.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.