Fast Cvvdp Implementation in C

New video-quality tool runs lean and mean—commenters fixate on 'poop' tests and Zig drama

TLDR: A lean C tool for judging video/image quality claims huge speed and memory wins, even single-threaded. Comments celebrate the efficiency but spiral into jokes about the “poop” benchmark, debates over fair testing, and confusion about needing Zig, splitting the crowd between believers and skeptics.

A new fast C build of the CVVDP metric—the thing that compares two images or videos and guesses how different they look to human eyes—just dropped, and the crowd went from curious to chaotic in seconds. Fans cheered the numbers: 91% less memory, 88% fewer CPU cycles, and ~18% faster even while running on a single CPU thread. Skeptics immediately side-eyed the benchmarks, which were (no joke) measured with a tool called “poop,” triggering a wave of giggles and memes. One commenter helpfully clarified that CVVDP aims to simulate human vision to spot differences, pointing to the arXiv work behind it.

Then the drama kicked in. Some argued it’s unfair to compare a single-threaded app to a multithreaded one—“turbo vs bicycle,” said one snark—while others claimed the real flex is doing more with less. The build instructions stirred their own storm: a C project that needs Zig? Cue the “is this C or Zig cosplay” jokes. The open-source Apache 2.0 license earned claps, and data nerds loved the JSON output for dashboards. Meanwhile, old-school purists asked whether classic scores (like simpler “how sharp is it” tests) still matter, and the thread devolved into poop puns faster than the benchmarks themselves. In short: performance bragging, tool-name comedy, and a community split between hype and healthy skepticism.

Key Points

  • fcvvdp is a fast C implementation of the CVVDP perceptual quality metric from the University of Cambridge.
  • Benchmarks on Linux (Core i7‑13700K) using “poop” show fcvvdp vs. cvvdp performance and resource use.
  • fcvvdp uses ~91% less RAM, ~88% fewer CPU cycles, and is ~18% faster in wall time; user time is ~15x more efficient.
  • Current limitation: fcvvdp runs single-threaded, while the baseline cvvdp uses multiple threads.
  • Build requires zlib‑rs, libunwind, Zig 0.15.x; provides CLI and library outputs; licensed under Apache 2.0 and derived from Vship (MIT).

Hottest takes

"simulate the human visual system to predict the perceived difference" — password4321
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.