Skip to content

SAHI 0.12.2 - Postprocessing Memory & Speed Fix 🩹

Choose a tag to compare

View all tags
@onuralpszr onuralpszr released this 20 Jul 19:10
0.12.2
6b2c4b9
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

A patch release fixing out-of-memory failures and severe slowdowns when postprocessing large prediction sets, plus model, packaging and documentation fixes gathered since 0.12.1.

🐛 Fixes

  • Postprocessing no longer allocates an N x N matrix for large inputs (#1395). v0.11 used a shapely STRtree and only compared nearby boxes; v0.12 replaced it with a dense matrix that is O(N^2) in time and memory regardless of layout. The TorchVision backend had no size guard and built seven N x N tensors on the GPU, so it ran out of memory first. The greedy loops only ever use matrix >= match_threshold, never the metric values, so the thresholded adjacency is now built directly from the pairs an STRtree reports as intersecting and stored as CSR. Fixes #1374.

  • RF-DETR local models can be loaded by class name, with corrected in docs (#1394).

  • MMDetection has_mask now handles RepeatDataset
    (#1387).

  • Dependency and minimum-version checks are enforced in import_utils
    (#1377).

  • All OpenCV distributions are kept on one version, avoiding conflicting
    installs (#1393).

📚 Documentation

  • Chinese translations updated and completed (#1371).
  • Docs default to YOLO26 and link the Ultralytics YOLO26 page (#1386), and YOLO26 is listed among the CLI models (#1378).

🧹 Maintenance & CI

  • Bumped astral-sh/setup-uv 8.2.0 → 8.3.2
    (#1389) and updated
    actions/checkout / actions/cache
    (#1385).
  • Relaxed the build requirement to >=0.10,<1.6
    (#1382).

⚡ Performance

33337 boxes, IOS metric, threshold 0.3. Same outputs before and after
(6571 predictions for greedy_nmm, 4639 for nmm).

CPU (Intel Core i7-13850HX):

backend greedy_nmm before after nmm before after
numpy 15.88s 0.18s 27.06s 0.21s
torchvision 6.22s 0.18s 17.34s 0.21s
numba 1.53s 1.55s 18.56s 0.22s

CUDA (RTX 4000 Ada Laptop, 12 GB):

backend greedy_nmm before after nmm before after
numpy 15.93s 0.26s 27.00s 0.21s
torchvision OOM, 4.14 GiB 0.17s OOM, 4.14 GiB 0.23s
numba 1.09s 1.09s 18.82s 0.22s

Inputs below 2000 boxes, and any non-positive threshold, stay on the dense path where it is faster.

Full Changelog:
0.12.1...0.12.2

Assets 2
Loading