The paper "Compact 3D Gaussian Splatting for Dense Visual SLAM" addresses the issue of high memory and storage costs in existing 3D Gaussian-based SLAM systems. The authors observe that the Gaussian ellipsoids created by these systems exhibit significant geometric similarities, leading to redundant representations. To mitigate this, they propose a compact 3D Gaussian Splatting SLAM system that reduces the number and parameter size of Gaussian ellipsoids. The key contributions include: 1. **Sliding Window-Based Masking**: This method removes redundant Gaussian ellipsoids by considering both volume and opacity, achieving faster rendering speed and efficient memory usage. 2. **Geometry Codebook and Quantization**: A novel codebook-based quantization method is introduced to compress the geometry attributes (scale and rotation) of Gaussian ellipsoids, reducing computational complexity and memory usage. 3. **Local-to-Global Bundle Adjustment**: A robust and accurate pose estimation method is proposed using a local-to-global bundle adjustment with reprojection loss, enhancing camera tracking accuracy. Experiments on various datasets demonstrate that the proposed method achieves faster training and rendering speeds (226% increase in rendering speed), lower memory usage (2.21× compression), and maintains state-of-the-art (SOTA) quality in scene representation. The method is also evaluated on different datasets, showing improved performance in camera pose estimation and 3D Gaussian reconstruction. The authors conclude that their compact 3D Gaussian SLAM system provides a comprehensive solution for dense visual SLAM, offering high-fidelity performance, fast training, and real-time rendering.The paper "Compact 3D Gaussian Splatting for Dense Visual SLAM" addresses the issue of high memory and storage costs in existing 3D Gaussian-based SLAM systems. The authors observe that the Gaussian ellipsoids created by these systems exhibit significant geometric similarities, leading to redundant representations. To mitigate this, they propose a compact 3D Gaussian Splatting SLAM system that reduces the number and parameter size of Gaussian ellipsoids. The key contributions include: 1. **Sliding Window-Based Masking**: This method removes redundant Gaussian ellipsoids by considering both volume and opacity, achieving faster rendering speed and efficient memory usage. 2. **Geometry Codebook and Quantization**: A novel codebook-based quantization method is introduced to compress the geometry attributes (scale and rotation) of Gaussian ellipsoids, reducing computational complexity and memory usage. 3. **Local-to-Global Bundle Adjustment**: A robust and accurate pose estimation method is proposed using a local-to-global bundle adjustment with reprojection loss, enhancing camera tracking accuracy. Experiments on various datasets demonstrate that the proposed method achieves faster training and rendering speeds (226% increase in rendering speed), lower memory usage (2.21× compression), and maintains state-of-the-art (SOTA) quality in scene representation. The method is also evaluated on different datasets, showing improved performance in camera pose estimation and 3D Gaussian reconstruction. The authors conclude that their compact 3D Gaussian SLAM system provides a comprehensive solution for dense visual SLAM, offering high-fidelity performance, fast training, and real-time rendering.