Gaussian Splatting Papers Guide

Research Landscape

Category Map

Tracking New Papers

import requests
from datetime import datetime, timedelta

# Search arXiv for recent 3DGS papers
def search_3dgs_papers(days_back=7):
    """Find recent 3D Gaussian Splatting papers on arXiv."""
    import arxiv

    query = (
        "ti:gaussian splatting OR "
        "abs:3D gaussian splatting OR "
        "abs:3DGS"
    )

    search = arxiv.Search(
        query=query,
        max_results=50,
        sort_by=arxiv.SortCriterion.SubmittedDate,
    )

    cutoff = datetime.now() - timedelta(days=days_back)
    papers = []
    for result in search.results():
        if result.published.replace(tzinfo=None) > cutoff:
            papers.append({
                "title": result.title,
                "authors": [a.name for a in result.authors[:3]],
                "url": result.entry_id,
                "published": result.published.strftime("%Y-%m-%d"),
                "categories": result.categories,
            })
    return papers

recent = search_3dgs_papers(days_back=14)
for p in recent:
    print(f"[{p['published']}] {p['title']}")
    print(f"  {', '.join(p['authors'])} | {p['url']}")

Key Methods Comparison

# Performance comparison (from original benchmarks)
methods = {
    "NeRF": {"psnr": 31.01, "fps": 0.03, "train_time": "hours"},
    "Instant-NGP": {"psnr": 33.18, "fps": 9.43, "train_time": "5 min"},
    "3DGS": {"psnr": 33.31, "fps": 134, "train_time": "6 min"},
    "Mip-Splatting": {"psnr": 33.46, "fps": 120, "train_time": "7 min"},
}

print(f"{'Method':<16} {'PSNR':>6} {'FPS':>8} {'Training':>10}")
print("-" * 44)
for name, m in methods.items():
    print(f"{name:<16} {m['psnr']:>6.2f} {m['fps']:>8.2f} "
          f"{m['train_time']:>10}")

Implementation Resources

# Original implementation
git clone https://github.com/graphdeco-inria/gaussian-splatting
cd gaussian-splatting
pip install -r requirements.txt

# Train on custom scene
python train.py -s path/to/colmap/data

# Real-time viewer
./SIBR_viewers/bin/SIBR_gaussianViewer_app \
  -m output/trained_model

Survey Papers

1. "A Survey on 3D Gaussian Splatting" (Chen et al., 2024) — comprehensive taxonomy
2. "3DGS: Recent Developments and Applications" (Wu et al., 2024) — application-focused
3. "Gaussian Splatting: A Survey" (Fei et al., 2024) — technical deep dive

Use Cases

1. Novel view synthesis: Photo-realistic rendering from sparse views
2. Real-time visualization: Interactive 3D scene exploration
3. Digital twins: Rapid scene reconstruction for simulation
4. VR/AR content: Real-time immersive experiences
5. Autonomous driving: Street-level scene understanding

References

• awesome-3D-gaussian-splatting
• Original 3DGS
• 3DGS Papers Collection