# New Release: Ultralytics v8.4.190 **URL:** \ **Category:** Discussion\ **Tags:** releases, announcements, ultralytics-official\ **Created:** [October 27, 2026, 1:14pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-190/2285 "2026-10-27T13:14:12Z")\ **Posts on this page:** 1\ **Page:** 1
**Author:** ![glenn-jocher](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/glenn-jocher/32/83_2.png) [@glenn-jocher](https://community.ultralytics.com/u/glenn-jocher)\ **Post date:** [October 27, 2026, 1:14pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-190/2285/1 "2026-10-27T13:14:12Z")
# 🚀 Ultralytics v8.4.190: Smarter YOLO26 Segmentation Training **Ultralytics YOLO v8.4.190 is here!** This release introduces mask-aware label assignment for YOLO26 segmentation training, adds RKNN INT8 export for depth models, and improves CoreML reliability. You’ll also find faster segmentation validation, better batch sizing, and fixes across inference, data handling, and tooling. ## 🎭 New Features - **Mask-aware YOLO26 segmentation training:** [PR #26773](https://github.com/ultralytics/ultralytics/pull/26773) by [@Bovey0809](https://github.com/Bovey0809) makes positive training matches consider both box and mask overlap. Bounded mask decoding and safer precision help control memory use and avoid numerical overflow. Other segmentation models and inference/export behavior remain unchanged. - **RKNN INT8 export for depth models:** [PR #26885](https://github.com/ultralytics/ultralytics/pull/26885) by [@Bovey0809](https://github.com/Bovey0809) enables INT8 export with hybrid precision that keeps error-prone layers in FP16. For calibration, use paired depth maps covering your deployment scene’s depth range. ## ⚡ Performance and Training Improvements - **Faster segmentation validation:** [PR #26898](https://github.com/ultralytics/ultralytics/pull/26898) by [@raimbekovm](https://github.com/raimbekovm) vectorizes COCO mask RLE encoding, reducing encoding time in tested workloads. - **OpenVINO CPU prefetching:** [PR #26897](https://github.com/ultralytics/ultralytics/pull/26897) by [@var1914](https://github.com/var1914) allows upcoming image batches to load while the current batch runs. - **More accurate AutoBatch sizing:** [PR #26894](https://github.com/ultralytics/ultralytics/pull/26894) by [@raimbekovm](https://github.com/raimbekovm) fixes memory accounting so batch selection better reflects GPU memory available to the process. - **Earlier validation feedback:** [PR #26891](https://github.com/ultralytics/ultralytics/pull/26891) by [@raimbekovm](https://github.com/raimbekovm) rejects unsupported semantic `single_cls` configurations before scanning masks. ## 📱 Export and Inference Fixes - **No duplicate CoreML segmentation detections:** [PR #26862](https://github.com/ultralytics/ultralytics/pull/26862) by [@amanharshx](https://github.com/amanharshx) fixes duplicate detections in segmentation exports with NMS. - **Safer CoreML inference on MPS:** [PR #26896](https://github.com/ultralytics/ultralytics/pull/26896) by [@amanharshx](https://github.com/amanharshx) fixes crashes caused by float64 outputs. - **Fewer dependency disruptions:** [PR #26863](https://github.com/ultralytics/ultralytics/pull/26863) by [@amanharshx](https://github.com/amanharshx) avoids unnecessary NumPy downgrades during CoreML inference. - **Improved oriented-box NMS precision:** [PR #26823](https://github.com/ultralytics/ultralytics/pull/26823) by [@wpyaoxueros](https://github.com/wpyaoxueros) fixes FP16 precision issues in embedded OBB NMS. ## 🗂 Data, Usability, and Tooling - **Correct handling of unlabeled classification images:** [PR #26865](https://github.com/ultralytics/ultralytics/pull/26865) by [@reubenjrouse](https://github.com/reubenjrouse) makes NDJSON conversion skip unlabeled images instead of assigning them to class 0. - **More robust FastSAM prompting:** [PR #26830](https://github.com/ultralytics/ultralytics/pull/26830) by [@aswanth-07](https://github.com/aswanth-07) handles text prompts when no masks qualify. - **Floating-point region coordinates:** [PR #26895](https://github.com/ultralytics/ultralytics/pull/26895) by [@Rahulbiradar9](https://github.com/Rahulbiradar9) fixes a `SolutionAnnotator` crash when drawing regions with floating-point coordinates. - **OCR-ready CPU runner:** [PR #26893](https://github.com/ultralytics/ultralytics/pull/26893) by [@glenn-jocher](https://github.com/glenn-jocher) adds Tesseract OCR with English data for scanned-PDF workflows. - **CUDA test configuration fix:** [PR #26900](https://github.com/ultralytics/ultralytics/pull/26900) by [@fcakyon](https://github.com/fcakyon) parses `YOLO_AUTODEVICE_FRACTION_FREE` as a float. ## 📚 Documentation Updates - **Clearer Ultralytics Platform guidance:** [PR #26888](https://github.com/ultralytics/ultralytics/pull/26888) by [@raimbekovm](https://github.com/raimbekovm) clarifies partial dataset and project cloning, prediction examples, providers, and the mobile import link. - **Better VLM discovery:** [PR #26884](https://github.com/ultralytics/ultralytics/pull/26884) by [@glenn-jocher](https://github.com/glenn-jocher) connects VLM documentation to the VLM Leaderboard and Platform model pages. ## 🙌 Thank You, Contributors! Thanks to everyone who helped ship this release! A special welcome to **@var1914, @reubenjrouse, and @wpyaoxueros** for their first contributions. ## 🛠 Try It and Share Your Feedback Upgrade to the latest release: ```bash pip install -U ultralytics ``` Explore the [v8.4.190 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.190) for release details, or browse the [full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.189...v8.4.190) to review all changes. **How does v8.4.190 work on your models and hardware?** Share your results, questions, and any issues below—especially if you’re training YOLO26 segmentation models or deploying with CoreML, RKNN, or OpenVINO. 🚀