# New Release: Ultralytics v8.4.193

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-193/2288>\
**Category:** Discussion\
**Tags:** ultralytics-official, releases, announcements\
**Created:** [October 30, 2026, 1:06pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-193/2288 "2026-10-30T13:06:53Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**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 30, 2026, 1:06pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-193/2288/1 "2026-10-30T13:06:53Z")

</div>

# 🚀 Ultralytics v8.4.193: Faster CPU Inference and More Reliable Workflows

**Ultralytics YOLO v8.4.193 is here!** This release delivers substantially faster CPU inference on some systems with ONNX Runtime and OpenVINO, smarter defaults for dynamic exports, and fixes across datasets, training, metrics, and visualization.

## ⚡ Headline Improvement: Faster CPU Inference

CPU ONNX Runtime and OpenVINO models now limit PyTorch preprocessing and postprocessing to **at most two threads** , reducing competition with the inference runtime.

- **On an 8-vCPU test machine:** prediction was **19–54% faster** across tested backends and batch sizes.
- **On 4 vCPUs:** results were mostly unchanged, with one ONNX Runtime batch test slightly slower.
- **PyTorch `.pt` setup is unchanged.**

In [PR #26942](https://github.com/ultralytics/ultralytics/pull/26942), [@var1914](https://github.com/var1914) contributed this targeted CPU performance improvement.

## 📦 Export and Installation Improvements

- **Dynamic export resolution and video reads:** [@Nikhi00718](https://github.com/Nikhi00718) updated behavior in [PR #26939](https://github.com/ultralytics/ultralytics/pull/26939) so prediction and validation default to the export metadata’s `imgsz` when no explicit size is supplied. Explicit settings still take precedence, and strided video reads now start at frame 0.
- **More reliable path checks:** [@MohammadHijjawi97](https://github.com/MohammadHijjawi97) fixed extensionless paths and dotted parent directories in [PR #26918](https://github.com/ultralytics/ultralytics/pull/26918) to improve suffix handling.
- **Runtime and solution fixes:** [@cainiao33](https://github.com/cainiao33) addressed ONNX Runtime variant reinstalls, VisionEye floating-point coordinates, and TrackZone grayscale frames in [PR #26932](https://github.com/ultralytics/ultralytics/pull/26932) for more consistent behavior.
- **Faster dependency installation:** [@JaviChulvi](https://github.com/JaviChulvi) updated automatic dependency installation in [PR #26950](https://github.com/ultralytics/ultralytics/pull/26950) to prefer `uv` when available on all platforms, including ARM64.

## 🧰 Dataset and Visualization Fixes

- **Individual image paths:** [@Vaishnavi220506](https://github.com/Vaishnavi220506) added support in [PR #26946](https://github.com/ultralytics/ultralytics/pull/26946) for individual image paths in dataset `img_path`.
- **Cleaner auto-annotation output:** [@raimbekovm](https://github.com/raimbekovm) fixed stale labels in [PR #26947](https://github.com/ultralytics/ultralytics/pull/26947) by replacing previous labels with empty files when no objects are detected.
- **Correct annotated-image splitting:** [@aswanth-07](https://github.com/aswanth-07) repaired relative-path label lookup in [PR #26948](https://github.com/ultralytics/ultralytics/pull/26948) for `autosplit` with `annotated_only`.
- **Better DOTA tiling:** [@raimbekovm](https://github.com/raimbekovm) changed label handling in [PR #26938](https://github.com/ultralytics/ultralytics/pull/26938) to clip labels to tile boundaries instead of dropping affected tiles.
- **No huge perspective ghost boxes:** [@8rulerstar](https://github.com/8rulerstar) updated `RandomPerspective` in [PR #26934](https://github.com/ultralytics/ultralytics/pull/26934) to discard objects crossing the projection plane.
- **Safer batch plotting:** [@Vaishnavi220506](https://github.com/Vaishnavi220506) fixed plotting in [PR #26925](https://github.com/ultralytics/ultralytics/pull/26925) to scale a copy rather than modify shared training or validation images.
- **Consistent plotting types:** [@mottopanikeiku](https://github.com/mottopanikeiku) aligned `Results.plot` typing in [PR #26924](https://github.com/ultralytics/ultralytics/pull/26924) so `line_width` is an integer, matching the configuration and annotator.

## ✅ Model, Metric, and Training Reliability

- **Correct class ordering:** [@raimbekovm](https://github.com/raimbekovm) updated `check_class_names` in [PR #26927](https://github.com/ultralytics/ultralytics/pull/26927) to return names in class-ID order.
- **Accurate semantic metrics:** [@aswanth-07](https://github.com/aswanth-07) fixed confusion-index overflow in [PR #26921](https://github.com/ultralytics/ultralytics/pull/26921) when using compact `uint8` labels.
- **More reliable W&B curves:** [@raimbekovm](https://github.com/raimbekovm) corrected class names and skipped label-less validation splits in [PR #26922](https://github.com/ultralytics/ultralytics/pull/26922) to keep class curves meaningful.
- **Clearer YOLOE errors:** [@raimbekovm](https://github.com/raimbekovm) added an explicit rejection in [PR #26920](https://github.com/ultralytics/ultralytics/pull/26920) for visual prompts on models with fused class embeddings.
- **More representative AutoBatch profiling:** [@raimbekovm](https://github.com/raimbekovm) updated profiling in [PR #26913](https://github.com/ultralytics/ultralytics/pull/26913) to use the real training loss for RT-DETR and semantic segmentation.
- **Distillation checkpoint recovery:** [@raimbekovm](https://github.com/raimbekovm) improved checkpoint loading in [PR #26937](https://github.com/ultralytics/ultralytics/pull/26937) so mid-run checkpoints can load their student model while resumed training retains wrapper state.
- **Safe loading for documented transforms:** [@Echoslayer](https://github.com/Echoslayer) enabled the documented letterbox classification transforms in [PR #26949](https://github.com/ultralytics/ultralytics/pull/26949) under safe loading.
- **Updated compatibility coverage:** [@glenn-jocher](https://github.com/glenn-jocher) refreshed slow CI in [PR #26951](https://github.com/ultralytics/ultralytics/pull/26951) for PyTorch 2.14 and 2.15, including newer-version coverage on Python 3.14.

## 📚 Documentation and Ultralytics Platform Updates

- [@glenn-jocher](https://github.com/glenn-jocher) added eligible Enterprise HIPAA BAA information and sticky project controls in [PR #26933](https://github.com/ultralytics/ultralytics/pull/26933) to clarify Platform capabilities.
- [@glenn-jocher](https://github.com/glenn-jocher) removed an outdated detect-only RKNN INT8 restriction in [PR #26945](https://github.com/ultralytics/ultralytics/pull/26945) from Platform export documentation.
- [@raimbekovm](https://github.com/raimbekovm) documented the `train` split argument, early-access annotation, and bulk trash controls in [PR #26923](https://github.com/ultralytics/ultralytics/pull/26923) for clearer workflow guidance.
- [@raimbekovm](https://github.com/raimbekovm) corrected integration export guides in [PR #26944](https://github.com/ultralytics/ultralytics/pull/26944), including format limitations and the transition from TF.js export to LiteRT.
- [@Echoslayer](https://github.com/Echoslayer) clarified `weights_dir` lookup and download behavior in [PR #26935](https://github.com/ultralytics/ultralytics/pull/26935) to make model-file handling easier to understand.

## 🛠 Upgrade and Share Your Results

Install this release with:

```bash
pip install --upgrade ultralytics==8.4.193

```

Explore the [v8.4.193 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.193) for the official announcement, or review the [full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.192...v8.4.193) to compare every change.

**Give it a try and tell us how it performs!** If you benchmark CPU inference, share your CPU, backend, batch size, and before-and-after timings below. Feedback on exports, training, and dataset workflows is welcome too.

🙏 Thanks to every contributor who helped make this release faster and more reliable!