Everything on this page is available under CC BY 4.0. Use it for anything, including commercial work. The only condition is attribution: say where the numbers came from and link back to this page.
If you cite one of these, link to this page rather than to the JSON file. That way a reader can find the method, the sample size and the caveats.
JPEG encoder presets at quality 90
Five encoding presets compared on identical source images at the same quality setting.
Quantisation tables across the quality range
The same two encoders at quality 40, 50, 60, 70 and 90, to find where the size curves cross.
Lossless PNG recompression on 261 files
Real PNG files recompressed with oxipng, with a pixel-by-pixel comparison against the original.
Image metadata across 1,086 JPEGs
How many real JPEG files carry EXIF, XMP, ICC profiles and usable GPS coordinates, and how large the metadata blocks are.
WebP round trip cost at quality 90
Each image encoded straight to JPEG versus through WebP and back, so the difference is the cost of the detour.
Re-encoding a damaged JPEG at high quality
Files first compressed at quality 45, then re-saved at 90, measured against both the clean path and the damaged one.
These are real measurements, not benchmarks: one collection of images, one encoder build, one machine. The articles describe what that does and does not generalise to.