[Release] Gaussian-LiteSplat v0.1.0 — Minimal, CPU-Friendly Gaussian Splatting Framework for Research & Prototyping

[abhaskumarsinha]

[Release] Gaussian-LiteSplat v0.1.0 — Minimal, CPU-Friendly Gaussian Splatting Framework for Research & Prototyping

Hey folks 👋

Just released Gaussian-LiteSplat — a lightweight and open-source framework for 3D Gaussian Splatting that runs on CPU and Google Colab (no CUDA needed!).

It’s a simplified implementation aimed at researchers, students, and hobbyists who want to experiment with COLMAP scenes, view synthesis, and efficient 3D reconstruction — without GPU headaches.

---

✨ Highlights

• 🚀 Runs on CPU / Colab
• 🧩 Supports SIMPLE_PINHOLE, PINHOLE, SIMPLE_RADIAL (COLMAP)
• 🎨 Trainable RGB colors (simplified from original paper)
• 🧠 Train 2K+ Gaussians within minutes
• 🔬 Great for small-scale 3D research, projection, and quick prototyping

---

⚙️ Install

!pip install git+https://github.com/abhaskumarsinha/Gaussian-LiteSplat.git

or

!git clone https://github.com/abhaskumarsinha/Gaussian-LiteSplat.git
%cd Gaussian-LiteSplat
!pip install -r requirements.txt

---

📸 Example

!python ./scripts/train_colmap.py \
    --colmap_scene '[COLMAP export folder]' \
    --litesplat_scene '[save folder]' \
    --output_dir 'output' \
    --total_gaussians 2200

---

📓 Example notebooks in /notebooks
📚 Repo: https://github.com/abhaskumarsinha/Gaussian-LiteSplat
🧑‍💻 Author: Abhas Kumar Sinha, 2025

---

🧾 Citation

@software{GaussianLiteSplat2025,
  author = {Abhas Kumar Sinha},
  title = {Gaussian-LiteSplat: A Simplified Gaussian Splatting Framework},
  year = {2025},
  url = {https://github.com/abhaskumarsinha/Gaussian-LiteSplat}
}

---

💬 Perfect For:

• Low-resource 3D research
• Teaching & visualization
• Prototyping Gaussian splatting without GPUs

Happy splatting 💫

Imported COLMAP Scene (Before Training)	Trained Gaussian LiteSplat Scene (After Training)