test(despeckle): add an op-level micro-benchmark for the page cleaner#37
Merged
Conversation
despeckle dominates pdfbook conversions (~72% of conv), but nothing measured WHERE inside clean() the ~50ms/page/core goes. benchCleaner times each Leptonica primitive the cleaner composes — read, the four selectBySize shapes (incl. the inverted-page variant whose giant background component is rendered back), the 43x43 dilate, the 7x7 open, the boolean ops, both counting passes, the G4 write — plus clean() end-to-end, on a deterministic synthetic 600-dpi A5 page, and writes the table to despeckle/docs/cleaner-baseline.md. The committed baseline (174.9ms clean(), sigma row covers 92.5%): dilate 43x43 = 21.5%, selectBySize on the inverted page x2 = 25.3% (22.2ms vs 15.2ms on the normal page — the background-component re-render penalty, measured), metrics-only countConnComp x2 = 13.5%, page selectBySize x2 = 17.4%, open 7x7 = 7.0%. Every following optimization is judged against this table. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Nothing measured where inside
clean()the ~50ms/page/core goes — and the optimization candidates (DWA morphology, metrics-pass skipping, selection restructuring) target different ops. This adds the op-level measurement first, so every following claim is judged against a committed table.What
CleanerBenchmark(test sources, mirroringPipelineBenchmark): times each Leptonica primitive the cleaner composes plusclean()end-to-end on a deterministic synthetic 600-dpi A5 page (3496×4961, glyph columns + dust + isolated blots + pin-holes so all three passes have real work), writingdespeckle/docs/cleaner-baseline.md. Task:./gradlew :despeckle:infrastructure:benchCleaner(-Preps=N).The baseline (clean() = 174.9ms single-threaded; Σ row covers 92.5%)
Notable: the inverted-page selectBySize costs 22.2ms vs 15.2ms on the normal page — the giant-background-component re-render penalty is now a measured fact (it motivates the planned De Morgan flip), and the two metrics-only counting passes are 13.5% of the page.
🤖 Generated with Claude Code