Jirka Borovec
|
153cde5854
|
feat(compact_mask): add `resize()` method and benchmark stage (#2227)
* feat: add resize() method and benchmark stage
* perf: optimise resize() — vectorised coords, L3 direct RLE
* refactor: split _rle_resize and extract _resize_crop
* test: expand resize() tests for scaling and edge cases
* refactor: switch resize helpers to F-order (column-major) RLE
* fix: merge True/True RLE junctions in _rle_join_cols
* perf: vectorize _rle_scale_col RLE re-encoding
* test: add density-dispatch and parallel-path resize tests
* refactor: harden resize() threading and RLE invariants
* fix: accurate resize timing and exact nearest-neighbour parity
* type: add explicit numpy typing to ndarray declarations
---------
Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
|
2026-04-22 15:34:12 +02:00 |
|
Jirka Borovec
|
9b7099d3fc
|
feat: add CompactMask for memory-efficient crop-RLE mask storage (#2159)
Dense (N, H, W) bool masks cause OOM for aerial imagery (1000 objects x
4K image ~ 8.3 GB). CompactMask encodes each mask as a run-length
sequence of its bounding-box crop, reducing typical usage to ~2 MB.
* fix: correct bounding box coordinates in CompactMask doctests
* feat: implement memory-efficient IoU and NMS with CompactMask integration
* test: add extensive tests for CompactMask IoU, NMS, and InferenceSlicer integration
* feat(examples): add CompactMask demo and benchmark
* feat(examples): expand CompactMask benchmark with new stages and metrics
* feat(tests): add detailed CompactMask tests for NMM, centroids, holes, and segments
* feat(compact_mask): add repack(), fix merge perf, and add parity tests
* fix(masks): handle empty crops by defaulting centroid to (0, 0)
* feat(compact_mask): add `bbox_xyxy` property and improve type annotations
* feat(compact_mask): enhance `with_offset` for clipping and add tests
* docs(compact_mask): unwrap prose and add per-operation speedup analysis
* perf(compact_mask): fast path in with_offset avoids decode/re-encode
* fix(benchmark): count NMS mismatches and explain exact-vs-resize difference
* test(compact_mask): add 121 parametrised random-scenario parity tests
* refactor: rename single-char variables to descriptive names
* fix(nms): remove resize-to-640 approximation from mask_non_max_suppression
* docs(compact_mask): update README with fresh benchmark results
* refactor(benchmark): improve summary table logic, add CSV export
* docs(compact_mask): update README with revised benchmark speedups and cleanup
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
|
2026-04-13 11:10:12 +02:00 |