Accessible 3D Gaussian Splatting in Google Colab: A Lightweight and Reproducible Pipeline for Environment Reconstruction

Three-dimensional reconstruction underpins applications in robotics, XR, and cultural heritage, yet practical deployment often demands expert setup and high-end hardware. This paper presents a lightweight and reproducible implementation of 3D Gaussian Splatting tailored for Google Colab's free tier (NVIDIA T4). Unlike popular community notebooks that are currently non-functional, our artifact is a maintained, end-to-end notebook that prioritizes stability, simplicity, and accessibility over algorithmic novelty. Concretely, we enable headless reliability for COLMAP, reduce preprocessing time and memory pressure by combining temporal frame subsampling (2 fps) with COLMAP's sequential matching (bounded overlap), configure a free-tier-friendly training schedule (7,000 iterations), and automate packaging of the resulting model for download. The final workflow is linear (cell-based) and requires no additional UI layers, minimizing cold-start overhead and dependency issues in cloud VMs. The contribution is an engineering pathway that makes 3D Gaussian Splatting usable for classrooms, fieldwork, and applied projects, while remaining faithful to established evaluation protocols (PSNR, SSIM, LPIPS) for future quantitative reporting. We release the notebook and instructions to foster adoption and reproducibility.