New Release: Ultralytics v8.4.147

Discussion
, ,
1

:rocket: Ultralytics v8.4.147: Grayscale Preprocessing, Better SAM Generation, and Stronger Tooling

Ultralytics v8.4.147 is now available! This release expands Albumentations preprocessing to grayscale images, improves SAM mask generation, makes downloads and CI environments more reliable, and refreshes onboarding and Ultralytics Platform automation documentation.

Quick overview: Better grayscale workflows, cleaner SAM results, faster YOLO26 onboarding, and more dependable development infrastructure. No new model family or weights are introduced in this release.

:glowing_star: New Features

:framed_picture: Albumentations Support for Grayscale Images

Albumentations preprocessing now runs on images with either 1 channel for grayscale or 3 channels for color, while images with unsupported channel counts continue to be skipped.

This enables configured transforms for medical, industrial, document, and scientific grayscale imagery. The change was introduced in PR #26133 by @onuralpszr.

:bullseye: Improved SAM Mask Generation

SAM generation gains several useful improvements:

  • Added min_mask_region_area to remove small mask regions and holes.
  • Corrected crop prompt scaling to cover the complete resized crop.
  • Improved confidence filtering for consistent behavior between direct generate() calls and standard prediction.
  • Standardized source image shape handling across arrays and tensors.
  • Added documentation demonstrating min_mask_region_area.

These updates arrived through PR #26129 by @raimbekovm.

:high_voltage: Improvements

:shield: More Reliable Downloads

The safe_download() utility now instructs curl to fail on HTTP errors rather than saving an error page as though it were a valid downloaded file. Failed downloads are therefore surfaced immediately and more clearly.

See PR #26126 by @lakshanthad for the implementation.

:test_tube: Stronger CI and Runner Environments

This release improves automated testing and runner reliability:

:books: Documentation and Onboarding

:robot: Expanded Ultralytics Platform Automation Guidance

New documentation makes Platform automation more accessible:

  • The platform-cli agent skill and ul cloud workflows are documented in PR #26136 by @JaviChulvi.
  • Platform Agents documentation now covers visual workflows, conditions, dataset collection, Slack alerts, execution options, and run monitoring through PR #26135 by @glenn-jocher.
  • Ask AI Autotraining guidance now covers dataset exploration, annotation, training, model comparison, export, and deployment through PR #26137 by @JaviChulvi.

:chequered_flag: Faster First Prediction with YOLO26

The quickstart has been rewritten around installing Ultralytics and running a first prediction with pretrained YOLO26, the recommended model for new projects.

The updated guidance includes clearer examples for:

  • CLI and Python
  • Docker and Conda
  • Headless environments
  • Development installations
  • Persistent settings

Explore the changes in PR #26128 by @raimbekovm.

:movie_camera: Additional Documentation Resource

A new video resource was added to the documentation through PR #26124 by @RizwanMunawar.

:package: Get v8.4.147

Upgrade to the latest release with:

pip install -U ultralytics

You can review the official Ultralytics v8.4.147 release or inspect every change in the full v8.4.146 to v8.4.147 changelog.

Give the new release a try and share your feedback, benchmarks, or any issues you encounter with the Ultralytics community. Happy building! :rocket: