<![CDATA[Stories by Sascha Kirch on Medium]]> https://medium.com/@SaschaKirch?source=rss-5c38dace9d5e------2 https://cdn-images-1.medium.com/fit/c/150/150/1*ZX2bVmxAYpL_yUVj2dzRjw.png Stories by Sascha Kirch on Medium https://medium.com/@SaschaKirch?source=rss-5c38dace9d5e------2 Medium Thu, 05 Nov 2026 00:16:22 GMT <![CDATA[Making SAM3 8x Faster — What the Profiler Actually Showed]]>

What profiling a state-of-the-art segmentation model teaches you about running research code efficiently

Continue reading on AI Advances »

]]> https://aiadvances.org/we-made-sam3-8x-faster-for-production-heres-what-the-profiler-actually-showed-b3b05f7472cb?source=rss-5c38dace9d5e------2 https://medium.com/p/b3b05f7472cb Mon, 06 Apr 2026 18:42:50 GMT 2026-04-06T21:33:51.285Z <![CDATA[Inside the CUDA Kernel: The GPU Implementation of 3D Gaussian Splatting]]>

Bonus Part: How Flat Arrays, Radix Sort, and Shared Memory Turn Conceptual Tiling Into 100+ FPS Rendering

Continue reading on AI Advances »

]]> https://aiadvances.org/inside-the-cuda-kernel-the-gpu-implementation-of-3d-gaussian-splatting-74c3261ed721?source=rss-5c38dace9d5e------2 https://medium.com/p/74c3261ed721 Sun, 01 Mar 2026 21:47:31 GMT 2026-03-02T07:38:46.435Z <![CDATA[The Tricks That Make Production 3DGS Fast (Even If Ours Isn’t)]]>

Part 4 of Building a 3D Gaussian Splatting Renderer — Understanding the Optimizations That Make Real-Time Rendering Possible

Continue reading on AI Advances »

]]> https://aiadvances.org/the-tricks-that-make-production-3dgs-fast-even-if-ours-isnt-6cd9928ec307?source=rss-5c38dace9d5e------2 https://medium.com/p/6cd9928ec307 Sat, 14 Feb 2026 22:49:40 GMT 2026-03-02T07:37:08.357Z <![CDATA[Splat Your Own Gaussians: From Circles to Ellipses]]>

Part 3 of Building a 3D Gaussian Splatting Renderer —Covariance Matrices, Jacobians, and Proper 2D Projection

Continue reading on AI Advances »

]]> https://aiadvances.org/splat-your-own-gaussians-from-circles-to-ellipses-96b69f1e7e3f?source=rss-5c38dace9d5e------2 https://medium.com/p/96b69f1e7e3f Sat, 07 Feb 2026 23:25:40 GMT 2026-03-02T07:36:19.662Z <![CDATA[Circles Are Not Gaussians (But Let’s Pretend They Are)]]>

Part 2 of Building a 3D Gaussian Splatting Renderer — Circles, Colors, and Transparency

Continue reading on AI Advances »

]]> https://aiadvances.org/circles-are-not-gaussians-but-lets-pretend-they-are-7bcf6db6efb8?source=rss-5c38dace9d5e------2 https://medium.com/p/7bcf6db6efb8 Sun, 01 Feb 2026 12:56:45 GMT 2026-03-02T07:35:56.194Z <![CDATA[I Built the Slowest 3D Gaussian Splatting Renderer… On Purpose]]>

Building a deliberately slow, crystal-clear Python renderer to reveal the implementation details that papers skip and GPU kernels hide

Continue reading on AI Advances »

]]> https://aiadvances.org/i-built-the-slowest-3d-gaussian-splatting-renderer-on-purpose-a8170b90d9b4?source=rss-5c38dace9d5e------2 https://medium.com/p/a8170b90d9b4 Sun, 25 Jan 2026 00:21:09 GMT 2026-03-02T07:35:37.818Z <![CDATA[Still Avoiding einsum()? It’s Time to Fix That]]>

Build Intuition, Master the Syntax, and Apply einsum() with Real Examples and a Handy Cheat Sheet

Continue reading on AI Advances »

]]> https://aiadvances.org/still-avoiding-einsum-its-time-to-fix-that-6779be94c7ed?source=rss-5c38dace9d5e------2 https://medium.com/p/6779be94c7ed Mon, 19 May 2025 00:49:23 GMT 2025-05-19T10:46:59.345Z <![CDATA[Mastering NumPy: Manual Metadata Manipulation for Memory-Efficient Arrays]]>

Part 3: NumPy Like a Pro: A Deep Dive Into Arrays and Performance

Continue reading on AI Advances »

]]> https://aiadvances.org/mastering-numpy-manual-metadata-manipulation-for-memory-efficient-arrays-fc461f64074f?source=rss-5c38dace9d5e------2 https://medium.com/p/fc461f64074f Mon, 12 May 2025 01:28:29 GMT 2025-05-19T05:27:09.162Z <![CDATA[The Power of Views: How NumPy Avoids Copies and Saves Memory]]>

Part 2: NumPy Like a Pro: A Deep Dive Into Arrays and Performance

Continue reading on AI Advances »

]]> https://aiadvances.org/the-power-of-views-how-numpy-avoids-copies-and-saves-memory-4421bea71dd2?source=rss-5c38dace9d5e------2 https://medium.com/p/4421bea71dd2 Sun, 04 May 2025 22:38:45 GMT 2025-05-12T06:39:55.247Z <![CDATA[Why NumPy Arrays Are So Fast (And How They Really Work)]]>

Part 1: NumPy Like a Pro: A Deep Dive Into Arrays and Performance

Continue reading on AI Advances »

]]> https://aiadvances.org/why-numpy-arrays-are-so-fast-and-how-they-really-work-1e5b56c75f8b?source=rss-5c38dace9d5e------2 https://medium.com/p/1e5b56c75f8b Mon, 28 Apr 2025 16:46:01 GMT 2025-05-12T06:37:51.800Z