pypi / faster-coco-eval
faster-coco-eval
What changed in faster-coco-eval 1, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-18.
1 security1 new APIs and options13 performance14 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initfaster-coco-eval 129 changes
Security 1
Hardens native RLE and evaluator inputs.
New APIs and options 1
Adds nano and micro evaluation ranges for very small objects.
Performance 13
Avoids accumulation copies, with a reported 1.11x boost.
Parallelizes batch RLE operations, with a reported 2.80x boost.
Parallelizes category-area accumulation.
Vectorizes computeOks, with a reported 6.05x boost.
Skips empty IoU pair calculations.
Partitions ignore ordering linearly, with a reported 3.77x boost.
Parallelizes image evaluation, with a reported 2.17x boost.
Releases the GIL during IoU computation, with a reported 2.59x boost.
Optimizes native and annotation dispatch.
Hoists Python list length, with a reported 1.09x boost.
Parallelizes RLE IoU evaluation, with a reported 2.63x boost.
Reuses sorted detections across maxDets.
Addresses memory problems; the notes do not specify the particular change.
Fixes and other changes 14
Warns when annotation ID 0 ambiguously matches the sentinel.
Adds guards for decoded RLE slices.
Resolves evaluation regressions.
Hardens extra evaluation helpers.
Adds support for gloo all gather.
Hardens low-level COCO API edge cases.
Preserves matrices during normalization.
Hardens native evaluator boundaries.
Retains orphan annotations in indexes.
Addresses review findings in the Python evaluation API.
Corrects mismatch indexing in rleIou.
Fixes precision over-estimation caused by monotone PR-curve interpolation.
Raises a clear ValueError when IoU=0.50 is absent from iouThrs.
Fixes a type error in faster_eval_api.