Gaussian Splatting Training Pipeline - Setup Guide¶

New Automation Tools

For the end-to-end pipeline orchestrator, see Automation Pipeline. For batch processing of asset libraries, see Library Batch Capture.

This repository includes two CLI scripts for training outside Blender:

tools/train_gs_simple.py - simple training for one folder or batch folders
tools/gs_training_pipeline.py - pipeline with optional COLMAP and batching

Quick Start (Blender Addon Output)¶

If you exported transforms.json, you can train directly without COLMAP.

# Single capture
python tools/train_gs_simple.py ./gs_capture/MyObject

# Batch process all capture folders
python tools/train_gs_simple.py ./gs_capture --batch --output ./trained_models

Option A - Original 3D Gaussian Splatting¶

The default path is the original 3DGS repo. You must run the script in an environment where 3DGS dependencies are installed.

# Clone
git clone https://github.com/graphdeco-inria/gaussian-splatting.git
cd gaussian-splatting

# Create env (example)
conda create -n gaussian_splatting python=3.8
conda activate gaussian_splatting

# Install dependencies
pip install -r requirements.txt
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn

Run training from that environment:

conda activate gaussian_splatting
python /path/to/GS_Blender/tools/train_gs_simple.py /path/to/capture

You can also set a repo path:

python tools/train_gs_simple.py ./gs_capture/MyObject --gs-path /path/to/gaussian-splatting

Option B - Nerfstudio (splatfacto)¶

tools/train_gs_simple.py supports Nerfstudio with the --nerfstudio flag:

pip install nerfstudio
python tools/train_gs_simple.py ./gs_capture/MyObject --nerfstudio

Using the Pipeline Script¶

tools/gs_training_pipeline.py is for larger batches and optional COLMAP processing.

# Process folders with default settings
python tools/gs_training_pipeline.py --input ./captures --output ./trained

# Skip COLMAP (recommended for Blender output)
python tools/gs_training_pipeline.py --input ./captures --output ./trained --skip-colmap

# Recursive scan
python tools/gs_training_pipeline.py --input ./captures --output ./trained --recursive

You can also use a YAML config:

# pipeline_config.yaml
input_folders:
  - "./captures"

output_base: "./trained"

colmap:
  enabled: false
  skip_if_transforms_exist: true

training:
  implementation: "original"  # or "nerfstudio"
  iterations: 30000
  white_background: true

python tools/gs_training_pipeline.py --config pipeline_config.yaml

Note: implementation: "gsplat" is not implemented in the pipeline script and will log a warning.

COLMAP (Only for Real Photos)¶

Blender output already includes camera parameters if you export transforms.json. COLMAP is only required for real photographs or video frames.

Output Structure Reference¶

output/
  ObjectName/
    images/
      image_0000.png
      image_0001.png
    transforms.json
    sparse/0/           # optional (COLMAP)
      cameras.txt
      images.txt
      points3D.txt

Troubleshooting¶

"gaussian-splatting not found": clone the repo or use --gs-path
"nerfstudio not found": install nerfstudio or remove --nerfstudio
No images found: make sure your capture folder has images/ and images inside