What is the difference between standard AABB and Rotated YOLO-OBB?

Standard Axis-Aligned Bounding Boxes (AABB) are locked horizontally and vertically, capturing excess background when objects are angled. Rotated Oriented Bounding Boxes (OBB) rotate with the object using 4 corner coordinates (x1,y1, x2,y2, x3,y3, x4,y4), maximizing detection precision.

Which computer vision use cases require Rotated OBB?

Rotated bounding boxes are essential for drone aerial inspection, satellite Earth observation, robotics grasping, overhead retail cart tracking, and inclined text OCR.

Rotated Bounding Box (YOLO-OBB) Online Annotation Tool

Why Standard Axis-Aligned Boxes Fail on Angled Objects

In standard object detection (AABB), bounding boxes must remain parallel to the screen axes. When objects are tilted—such as a drone photo of ships at sea, diagonal parking lots, or angled circuit board components—a standard box captures mostly background noise.

Standard Upright (AABB)

Captures massive empty background when objects rotate at 45°. Causes overlapping bounding boxes and degrades model mAP precision.

Polytate Rotated (YOLO-OBB)

Hugs the exact contour of the rotated object using 4 corner vertices. Drastically reduces false positives in robotics and aerial inspection.

The YOLO-OBB Coordinate Format

Polytate automatically computes and normalizes the four corner points required by Ultralytics YOLOv8-OBB and YOLO11-OBB:

# YOLO-OBB 4-Corner Normalized Polygon Format

<class_id> <x1> <y1> <x2> <y2> <x3> <y3> <x4> <y4>

Coordinates start from the top-left vertex and progress clockwise around the rotated polygon.

Training YOLOv8-OBB with Polytate Datasets

from ultralytics import YOLO

# 1. Load pre-trained YOLO-OBB model
model = YOLO('yolov8n-obb.pt')

# 2. Train on the exported Polytate dataset
model.train(data='polytate_obb_dataset.yaml', epochs=100, imgsz=1024)

Built for Specialized Vision Workloads

🛰️ Satellite & Aerial Drone Vision

Detect ships, aircraft on runways, storage tanks, and agricultural fields with zero orientation bias.

🤖 Robotics Grasping & Warehouse Logistics

Provide robotic suction and mechanical grippers with the exact orientation angle of parcels and manufactured parts.