update gsplat summary

w-m · Sep 5, 2024 · 6bdcadb · 6bdcadb
1 parent 3c2a6f0
commit 6bdcadb
Showing 1 changed file with 8 additions and 3 deletions.
diff --git a/methods/hu2024gsplat.md b/methods/hu2024gsplat.md
@@ -1,8 +1,13 @@
 ### gsplat
 
-This approach combines 3D Gaussian Splatting as Markov Chain Monte Carlo <a target="_blank" href="https://ubc-vision.github.io/3dgs-mcmc/">(3DGS-MCMC)</a> with compression techniques from the 
+This approach leverages 3D Gaussian Splatting as Markov Chain Monte Carlo <a target="_blank" href="https://ubc-vision.github.io/3dgs-mcmc/">(3DGS-MCMC)</a>, interpreting the training process of positioning and optimizing Gaussians as a sampling procedure rather than minimizing a predefined loss function. 
 
-<span class="text-item"><span class="text-color-box" style="background-color: rgb(31, 119, 180);"></span><a href="#morgenstern2024compact" style="display: inline;">Self-Organizing Gaussians</a></span> 
+Additionally, it incorporates compression techniques derived from the 
+<span class="text-item"><span class="text-color-box" style="background-color: rgb(31, 119, 180);"></span><a href="#morgenstern2024compact" style="display: inline;">Self-Organizing Gaussians</a></span> paper, which organizes the parameters of 3DGS in a 2D grid, capitalizing on perceptual redundancies found in natural scenes, thereby significantly reducing storage requirements. 
 
-paper and <a target="_blank" href="https://aras-p.info/blog/2023/09/27/Making-Gaussian-Splats-more-smaller/">Making Gaussian Splats more smaller</a>. It is implemented in <a target="_blank" href="https://docs.gsplat.studio">gsplat</a>, an open-source library for CUDA-accelerated differentiable rasterization of 3D gaussians with Python bindings. The library is inspired by the SIGGRAPH paper "3D Gaussian Splatting for Real-Time Rendering of Radiance Fields", but gsplat is faster, more memory efficient, and with a growing list of new features.
+Further compression is achieved by applying methods from 
+<a target="_blank" href="https://aras-p.info/blog/2023/09/27/Making-Gaussian-Splats-more-smaller/">Making Gaussian Splats more smaller</a>, which reduces the size of Gaussian splats by clustering spherical harmonics into discrete elements and storing them as FP16 values.
+
+This technique is implemented in 
+<a target="_blank" href="https://docs.gsplat.studio">gsplat</a>, an open-source library designed for CUDA-accelerated  differentiable rasterization of 3D Gaussians, equipped with Python bindings.