Profile Linux OCR image processing and recognition throughput #11
Labels
No labels
status:done
status:in-progress
status:ready
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
rexlManu/garmith-app#11
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Two real OCR workers are active, but the small generated-image test showed 747 ms sequential versus 737 ms concurrent for eight images. Measure decode, resize, pixel transform, encoding and recognition separately with temporary profiles. Use results to choose a small optimization only if measured. Preserve text, coordinates and supported languages. No live account or desktop capture. Record live verification limits.
Completed in
6c2997a. The previous benchmark used incidental helper IPC, intentionally serialized on scheduler worker0, so it could not measure loot-pool concurrency. pnpm profile:ocr now separates helper batches, image processing stages and direct maintained-source OCR pool batches. Eight generated 430x250 images: sequential349-372ms, parallel193-214ms; 900x600: sequential718-763ms, parallel460-488ms. About1.6-1.8x for these inputs. Preprocessing averages8.6/28.6ms, with PNG encoding4.9/13.7ms. No speculative runtime optimization added. All text checks pass; methods and limits are in docs/ocr-performance.md. Live verification is #12, installed release update after that is #13.