Simple NanoGS GUI

Simple NanoGS GUI is a Windows-focused fork of the original NanoGS project for training-free Gaussian splat simplification.

Original upstream project: https://github.com/saliteta/NanoGS

Project page: https://saliteta.github.io/NanoGS/

This repository keeps the upstream simplification core, then adds a practical desktop workflow and format-conversion tooling for local batch use.

What changed compared to the original repo

• Added a Tkinter desktop GUI for single-file and folder-based batch processing.
• Added saved parameter profiles and persisted local GUI settings.
• Added support for .sog input by decoding it to a temporary .ply file in Python.
• Added .spz and .sog round-trip export support through gsbox.exe.
• Added a quality-test mode that generates about 40 nearby parameter variants and writes a CSV manifest.
• Added CPU/GPU selection for cost evaluation with automatic backend choice.
• Added automatic block_edges tuning plus a manual override for large scenes.
• Added benchmark scripts for backend and block-size tuning.
• Added a Windows launcher batch file for the GUI.
• Extended the CLI with --device and --block_edges.

Repository contents

• nanogs_gui.py: desktop GUI for running simplification jobs.
• simplification.py: CLI entry point and simplification pipeline.
• install_gpu_support.bat: Windows helper that installs CuPy into .venv for source-mode GPU execution.
• utils/sog_utils.py: .sog decoder and conversion to .ply.
• benchmark_block_edges.py: compare runtime across block sizes and devices.
• benchmark_device_threshold.py: compare CPU and GPU behavior across scene sizes.
• start_nanogs_gui.bat: Windows launcher that runs the GUI from .venv.

Requirements

• Python 3.10+ on Windows.
• Base dependencies listed in requirements.txt.
• Optional: a CUDA-capable GPU plus a matching CuPy build for GPU cost evaluation.
• Optional: gsbox.exe if you want .spz conversion or .sog/.spz export.

Example environment setup:

python -m venv .venv
.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

Optional GPU backend:

python -m pip install cupy-cuda12x

Use a CuPy package that matches your CUDA runtime.

On Windows, you can also run:

install_gpu_support.bat

That script creates .venv if needed, installs requirements.txt, prompts for the matching CuPy package, and verifies CUDA device access.

If you only need the CPU path, requirements.txt is sufficient.

GUI workflow

Start the GUI with either:

.venv\Scripts\python.exe nanogs_gui.py

or:

start_nanogs_gui.bat

The GUI supports:

• Single-file processing for .ply, .sog, and .spz.
• Folder processing with optional recursion.
• Custom output folders and suffixes.
• Parameter profiles for different compression presets.
• Automatic or explicit CPU/GPU execution.
• Quality-test sweeps around the current ratio, k, opacity threshold, and lam_sh.

Format handling

• .ply files are processed directly.
• .sog files are converted to temporary .ply files inside the app, then can be exported back to .sog with GSBox.
• .spz files are converted through GSBox. A fallback converter command can also be configured in the GUI.

gsbox.exe is not stored in this repository. Put it next to nanogs_gui.py or configure its path in the GUI.

CLI usage

The command-line flow is still available:

python simplification.py --ply scene.ply -r 0.3 --k 8 --opacity_threshold 0.18 --lam_geo 1.0 --lam_sh 0.5 --device auto --block_edges 0

Key fork-specific CLI additions:

• --device {auto,cpu,gpu} chooses the edge-cost backend.
• --block_edges N overrides the auto-tuned edge block size.

Benchmarks

Two helper scripts are included for tuning and regression checks:

python benchmark_block_edges.py
python benchmark_device_threshold.py

They expect local benchmark .ply files and are intended as developer utilities rather than end-user entry points.

Notes

• nanogs_gui_settings.json is generated locally and is intentionally ignored.
• Benchmark logs, virtual environments, and workspace files are also ignored.
• This repository is intended as a practical GUI fork of NanoGS rather than a mirror of the upstream paper repository.

Citation

If the original NanoGS project helps your work, please cite the paper in CITATION.bib.

License

See LICENSE in this repository.