Skip to content

docs: add roadmap ideas from comparing with JPEG XL vs AVIF for web use-cases - #59

Merged
nucliweb merged 1 commit into
mainfrom
docs/roadmap-ideas
Oct 19, 2026
Merged

nucliweb merged 1 commit into
mainfrom
docs/roadmap-ideas

Conversation

@nucliweb

Copy link
Copy Markdown
Owner

What

JPEG XL vs AVIF for web use-cases (Jake Archibald, 2026) and its benchmark, image-codec-bench, measure things this tool does not yet. ROADMAP.md gains them as ideas, not commitments:

  • More codecs: jixel, a second JPEG XL encoder (Rust, BSD-3-Clause) that the post finds much closer to AVIF than libjxl on photographs (+11–12% vs +41–50%); an opt-in variant like rav1e and SVT-AV1. No CLI, so the image would build a wrapper.
  • Encode memory: peak resident memory from the /usr/bin/time wrapper every encode already runs under (-v GNU, -l BSD).
  • Tool versions in the results: encoder, decoder and metric versions in the CSV and the report.
  • An independent second metric: optional ColorVideoVDP, outside the default image (Python + PyTorch, ~1 GB).
  • Decode with platform decoders: Apple ImageIO (sips) and browsers, plus decode time; the HEIC 4:2:0 lesson already shows a 1.5–8.4 point libheif vs ImageIO gap.
  • Demo material: an image with alpha; all seven demo images are opaque.

…se-cases

Six ideas drawn from Jake Archibald's comparison and its benchmark tool: jixel as a second JPEG XL encoder (under "More codecs"), encoder peak memory from the existing /usr/bin/time wrapper, tool versions in the CSV and report, ColorVideoVDP as an optional independent metric, decoding with platform decoders and timing it, and an image with alpha in the demo set.
@nucliweb
nucliweb merged commit df835ed into main Oct 19, 2026
1 check passed
@nucliweb
nucliweb deleted the docs/roadmap-ideas branch October 19, 2026 20:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant