--- license: mit pretty_name: Browser JPEG/WebP target-size benchmarks language: - en tags: - image-compression - jpeg - webp - canvas - benchmarks - web-performance configs: - config_name: default data_files: - split: train path: data/browser-encoder-benchmarks.csv --- # Browser JPEG/WebP target-size benchmarks Measured byte outputs of browser-side `canvas.toBlob()` encoding for **3 synthetic image classes** (photo, screenshot, flat) × **3 source resolutions** × **2 formats** (JPEG, WebP) × **6 target sizes** (20KB, 50KB, 100KB, 200KB, 500KB, 1MB) = **108 measurement runs**. Each row records the quality and scale parameters a binary search chose, the final output dimensions, and the resulting byte size — so you can see what it actually costs, in pixels and quality, to hit a target file size in the browser. ## What the data shows - A 12MP photo cannot keep its full resolution at a 200KB JPEG target: it lands at 3139×2354 (7.4MP) with quality already at the 0.40 floor. - Screenshots are far easier: 1920×1080 fits 200KB at quality 0.76, but the same image needs 395×222 to fit 20KB. - A near-flat image always encodes to ~14KB regardless of the target (20KB or 1MB) — there is nothing left to compress. Any tool claiming "compressed to 1MB" on such an image is reporting the target, not the result. - 32 of the 108 runs had to give up resolution; the other 76 fit by lowering quality alone. ## Files - `data/browser-encoder-benchmarks.csv` — the 108 runs. - `METHOD.md` — browser version, machine, how the synthetic images are generated, the search algorithm, and five stated limitations (synthetic images are not your photos, single browser, encoder non-monotonicity, ICC handling, WebP quality scale ≠ JPEG). ## Reproduce The harness that produced this CSV is public, and the synthetic images use a fixed seed, so a rerun on the same browser version yields the same numbers: - Source repository: - Archived dataset (DOI): The same target-size search runs in [ExactSize](https://crunchimage.com), a browser-only image compressor that never uploads your files. ## Licence MIT. Free to reuse with attribution to this dataset.