<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom"><title>Journal by Aruodore</title><id>https://journal.aruodore.com/</id><link href="https://journal.aruodore.com/feed.xml" rel="self"/><link href="https://journal.aruodore.com/"/><updated>2026-08-04T00:00:00Z</updated><author><name>Lucas Aruodore Adomi</name></author><entry><title>The geometry of image formation</title><id>blog/blog/geometry-of-image-formation.md</id><link href="https://journal.aruodore.com/blog/geometry-of-image-formation"/><published>2026-08-04T00:00:00Z</published><updated>2026-08-04T00:00:00Z</updated><summary>A practical derivation of the pinhole camera model, its coordinate systems, and the assumptions hidden inside a projection matrix.</summary><category term="computer-vision"/><category term="geometry"/></entry><entry><title>Normalize camera coordinates before blaming calibration</title><id>notebook/notebook/normalizing-camera-coordinates.md</id><link href="https://journal.aruodore.com/notebook/normalizing-camera-coordinates"/><published>2026-08-02T00:00:00Z</published><updated>2026-08-02T00:00:00Z</updated><summary>A pixel-space residual looked like calibration drift, but the comparison mixed normalized and distorted coordinates.</summary><category term="computer-vision"/><category term="geometry"/></entry><entry><title>A benchmark is an experimental record</title><id>blog/blog/reproducible-gpu-benchmarks.md</id><link href="https://journal.aruodore.com/blog/reproducible-gpu-benchmarks"/><published>2026-07-21T00:00:00Z</published><updated>2026-08-01T00:00:00Z</updated><summary>GPU timing numbers become useful only when hardware, synchronization, sampling, and raw observations remain attached to the result.</summary><category term="scientific-computing"/><category term="ai-engineering"/></entry><entry><title>GPU synchronization changed the ranking</title><id>notebook/notebook/gpu-synchronization-changed-the-result.md</id><link href="https://journal.aruodore.com/notebook/gpu-synchronization-changed-the-result"/><published>2026-07-18T00:00:00Z</published><updated>2026-07-19T00:00:00Z</updated><summary>Adding synchronization reversed an apparent latency advantage between two inference paths.</summary><category term="scientific-computing"/><category term="ai-engineering"/></entry><entry><title>Half precision made the solver look converged</title><id>notebook/notebook/failed-half-precision-solver.md</id><link href="https://journal.aruodore.com/notebook/failed-half-precision-solver"/><published>2026-06-29T00:00:00Z</published><updated>2026-06-29T00:00:00Z</updated><summary>A residual plateau was mistaken for convergence after reduced precision erased the update direction.</summary><category term="numerical-methods"/><category term="scientific-computing"/></entry><entry><title>Reading a method paper backwards</title><id>notebook/notebook/reading-a-paper-backwards.md</id><link href="https://journal.aruodore.com/notebook/reading-a-paper-backwards"/><published>2026-06-11T00:00:00Z</published><updated>2026-06-11T00:00:00Z</updated><summary>Starting with evaluation details and failure cases can expose a paper&apos;s actual contribution faster than reading from the abstract forward.</summary><category term="research-practice"/><category term="ai-engineering"/></entry></feed>