QSplat is a multiresolution point rendering system designed for large meshes. It uses a bounding sphere hierarchy for visibility culling, level-of-detail control, and rendering. The system employs a single data structure for view frustum culling, backface culling, level-of-detail selection, and rendering. It is compact, efficient, and suitable for large data sets. QSplat is implemented for a large-scale 3D digitization project and can quickly launch, maintain an interactive frame rate, and refine progressively to high-quality images. It has been tested on scanned models with hundreds of millions of samples. The system uses a hierarchy of bounding spheres for visibility culling, level-of-detail control, and rendering. Each node contains the sphere center and radius, a normal, the width of a normal cone, and optionally a color. The hierarchy is constructed as a preprocessing step and stored on disk. The rendering algorithm traverses the hierarchy, culling nodes not visible, and drawing splats when necessary. QSplat uses a simple rendering algorithm based on traversing a bounding sphere hierarchy, making it suitable for browsing models with 100 million to 1 billion samples. QSplat's preprocessing algorithm begins with a triangular mesh and assigns sphere sizes to input vertices. The algorithm builds the tree by splitting vertices along the longest axis of the bounding box, recursively computing subtrees, and finding the bounding sphere of the two children. The tree is stored in breadth-first order, and nodes are quantized to reduce storage requirements. QSplat's design decisions include quantization, file layout, splat shape, and the choice of splatting, all aimed at achieving fast rendering and compact representation. QSplat's splat shape is chosen to affect the quality of the final image. It uses a variety of splat kernels, including squares, circles, and Gaussians. The choice of splat shape can significantly impact the quality of the final image. QSplat's design also includes a point-based system, which is most suitable for scenes with uniformly-sized geometric detail. It is less effective for scenes with large, smooth regions but can still produce good visual quality. QSplat's performance is optimized for interactive rendering, with a goal of maintaining a user-settable interactive frame rate. It has been tested on various hardware, including low-end machines, and can render large models efficiently. QSplat's preprocessing time is significantly faster than many contemporary mesh decimation algorithms, making it suitable for large-scale visualization. QSplat's system has been compared to previous work in point rendering, visibility culling, level-of-detail control, and geometric compression. It has demonstrated real-time progressive rendering of large models and has potential applications in areas where 3D rendering was previously impractical. Future work includes incorporating additional techniques to improve time and space efficiency, as well as exploring the integration of QSplat with different kinds of algorithms within computer graphics.QSplat is a multiresolution point rendering system designed for large meshes. It uses a bounding sphere hierarchy for visibility culling, level-of-detail control, and rendering. The system employs a single data structure for view frustum culling, backface culling, level-of-detail selection, and rendering. It is compact, efficient, and suitable for large data sets. QSplat is implemented for a large-scale 3D digitization project and can quickly launch, maintain an interactive frame rate, and refine progressively to high-quality images. It has been tested on scanned models with hundreds of millions of samples. The system uses a hierarchy of bounding spheres for visibility culling, level-of-detail control, and rendering. Each node contains the sphere center and radius, a normal, the width of a normal cone, and optionally a color. The hierarchy is constructed as a preprocessing step and stored on disk. The rendering algorithm traverses the hierarchy, culling nodes not visible, and drawing splats when necessary. QSplat uses a simple rendering algorithm based on traversing a bounding sphere hierarchy, making it suitable for browsing models with 100 million to 1 billion samples. QSplat's preprocessing algorithm begins with a triangular mesh and assigns sphere sizes to input vertices. The algorithm builds the tree by splitting vertices along the longest axis of the bounding box, recursively computing subtrees, and finding the bounding sphere of the two children. The tree is stored in breadth-first order, and nodes are quantized to reduce storage requirements. QSplat's design decisions include quantization, file layout, splat shape, and the choice of splatting, all aimed at achieving fast rendering and compact representation. QSplat's splat shape is chosen to affect the quality of the final image. It uses a variety of splat kernels, including squares, circles, and Gaussians. The choice of splat shape can significantly impact the quality of the final image. QSplat's design also includes a point-based system, which is most suitable for scenes with uniformly-sized geometric detail. It is less effective for scenes with large, smooth regions but can still produce good visual quality. QSplat's performance is optimized for interactive rendering, with a goal of maintaining a user-settable interactive frame rate. It has been tested on various hardware, including low-end machines, and can render large models efficiently. QSplat's preprocessing time is significantly faster than many contemporary mesh decimation algorithms, making it suitable for large-scale visualization. QSplat's system has been compared to previous work in point rendering, visibility culling, level-of-detail control, and geometric compression. It has demonstrated real-time progressive rendering of large models and has potential applications in areas where 3D rendering was previously impractical. Future work includes incorporating additional techniques to improve time and space efficiency, as well as exploring the integration of QSplat with different kinds of algorithms within computer graphics.