Surfels are point primitives used for efficient rendering of complex geometric objects at interactive frame rates. Unlike traditional surface discretizations like triangles, surfels do not have explicit connectivity and store attributes such as depth, texture color, and normal. A pre-processing step creates an octree-based surfel representation of a geometric object. During rendering, a hierarchical forward warping algorithm projects surfels to a z-buffer, and visibility splatting determines visible surfels and holes. Visible surfels are shaded using texture filtering, Phong illumination, and environment mapping. Image reconstruction methods like supersampling offer flexible speed-quality trade-offs. The surfel rendering pipeline is suitable for low-cost, real-time graphics such as games due to its simplicity and efficiency. Surfels provide complex shape, low rendering cost, and high image quality. They are particularly effective for models with rich, organic shapes or high surface details. The surfel rendering pipeline complements existing graphics pipelines and is positioned between conventional geometry-based approaches and image-based rendering. It trades memory overhead for rendering performance and quality. Surfels are not well-suited for flat surfaces but work well for complex, organic shapes. The surfel rendering system uses a hierarchical data structure called the LDC tree, which allows efficient rendering and image reconstruction. The system includes an efficient hierarchical representation, high-quality texture filtering, accurate visibility calculations, and flexible speed-quality trade-offs. The surfel rendering pipeline can handle exceedingly complex models and is amenable to hardware implementation. The rendering pipeline involves sampling and surfel rendering. Sampling converts geometric objects and textures into surfels, using ray casting to create three orthogonal layered depth images (LDIs). The LDC tree is used for rendering, with hierarchical block warping and visibility testing. Texture filtering is performed during preprocessing and rendering, with prefiltered texture samples stored per surfel. Shading is done using per-surfel normals, and image reconstruction includes hole filling and antialiasing. The surfel rendering system offers several speed-quality trade-offs, with visibility splatting effectively detecting holes and improving image reconstruction. Supersampling is naturally integrated for antialiasing. The system demonstrates high image quality at interactive frame rates, and further optimization could enhance performance. The system supports hardware implementation and can be integrated with existing graphics pipelines. Future extensions include volume data, point clouds, and non-synthetic LDIs, as well as hardware architecture design for surfel rendering.Surfels are point primitives used for efficient rendering of complex geometric objects at interactive frame rates. Unlike traditional surface discretizations like triangles, surfels do not have explicit connectivity and store attributes such as depth, texture color, and normal. A pre-processing step creates an octree-based surfel representation of a geometric object. During rendering, a hierarchical forward warping algorithm projects surfels to a z-buffer, and visibility splatting determines visible surfels and holes. Visible surfels are shaded using texture filtering, Phong illumination, and environment mapping. Image reconstruction methods like supersampling offer flexible speed-quality trade-offs. The surfel rendering pipeline is suitable for low-cost, real-time graphics such as games due to its simplicity and efficiency. Surfels provide complex shape, low rendering cost, and high image quality. They are particularly effective for models with rich, organic shapes or high surface details. The surfel rendering pipeline complements existing graphics pipelines and is positioned between conventional geometry-based approaches and image-based rendering. It trades memory overhead for rendering performance and quality. Surfels are not well-suited for flat surfaces but work well for complex, organic shapes. The surfel rendering system uses a hierarchical data structure called the LDC tree, which allows efficient rendering and image reconstruction. The system includes an efficient hierarchical representation, high-quality texture filtering, accurate visibility calculations, and flexible speed-quality trade-offs. The surfel rendering pipeline can handle exceedingly complex models and is amenable to hardware implementation. The rendering pipeline involves sampling and surfel rendering. Sampling converts geometric objects and textures into surfels, using ray casting to create three orthogonal layered depth images (LDIs). The LDC tree is used for rendering, with hierarchical block warping and visibility testing. Texture filtering is performed during preprocessing and rendering, with prefiltered texture samples stored per surfel. Shading is done using per-surfel normals, and image reconstruction includes hole filling and antialiasing. The surfel rendering system offers several speed-quality trade-offs, with visibility splatting effectively detecting holes and improving image reconstruction. Supersampling is naturally integrated for antialiasing. The system demonstrates high image quality at interactive frame rates, and further optimization could enhance performance. The system supports hardware implementation and can be integrated with existing graphics pipelines. Future extensions include volume data, point clouds, and non-synthetic LDIs, as well as hardware architecture design for surfel rendering.