🚀 AlbumentationsX 2.0.17 — Native Oriented Bounding Boxes (OBB)
Hi everyone,
AlbumentationsX 2.0.17 introduces a major feature:
🚀 Native Oriented Bounding Box (OBB) Support
AlbumentationsX now supports both:
HBB — horizontal (axis-aligned) bounding boxes
OBB — oriented (rotated) bounding boxes
This means augmentation pipelines are now fully consistent for rotated object detection tasks such as:
Aerial imagery
Scene text detection
Industrial inspection
Autonomous driving
No hacks. No post-processing fixes. No geometry mismatches.
✔️ OBB Support Across 40+ Transforms
OBB is now supported in a wide range of spatial and geometric transforms, including:
…and many more.
Every transform can be explored interactively here:
https://explore.albumentations.ai
🧭 New Bounding Box Format: cxcywh (unnormalized)
We also added support for a new bounding box format:
cxcywh (unnormalized pixel coordinates)
This format:
Matches YOLO layout
Uses pixel coordinates
Is not normalized
Works naturally in center-based detection workflows
No more manual conversion when working with rotated detection pipelines.
📘 Example Notebook
We prepared a full example demonstrating OBB + affine transforms on aerial boat imagery:
Documentation
Comprehensive documentation for the Albumentations library
This notebook shows how rotated boxes behave under geometric transformations — and why proper OBB handling matters.
🔎 Full Release Notes
Complete technical details, full transform list, and implementation notes:
Release 🛠 Albumentations 2.0.17 Release Notes · albumentations-team/AlbumentationsX · GitHub
📦 AlbumentationsX 2.0.17 🚀 Major Feature: Oriented Bounding Boxes (OBB) Support This release introduces native Oriented Bounding Box (OBB) support across a wide range of transforms. AlbumentationsX...
Why This Matters
Rotated object detection is becoming standard in:
Aerial and satellite imagery
OCR and scene text
Industrial QA
Robotics and autonomy
Without correct OBB handling, augmentations silently introduce label errors.
With AlbumentationsX 2.0.17, rotated detection pipelines become first-class citizens.
As always, thank you for building with Albumentations.
More updates coming soon.
Vladimir
Founder and CEO of Albumentations.AI