YOLO Ecosystem
"You Only Look Once" — fast single-stage object detectors, their versions, metrics and deployment.
This is an overview — pick a sub-topic below.
Sub-topics (13)
YOLO
A single network predicts boxes and classes in one pass over the image — fast enough for real-time video.
YOLO Versions (v3 to YOLO12)
From Darknet's YOLOv3/v4 to Ultralytics YOLOv5/v8/YOLO11 and newer research variants.
Ultralytics
The Python package for training, validating, tracking and exporting YOLO models.
YOLO-World and Open-Vocabulary Detection
Detect objects described by arbitrary text prompts, without retraining.
Darknet
The original C framework behind YOLOv1–v4.
Non-Maximum Suppression
Remove duplicate overlapping boxes, keeping the most confident one.
Anchor-Based vs Anchor-Free Detection
Predict offsets from preset anchor boxes, or predict object centres/edges directly.
Bounding-Box Regression and IoU Losses
Learn box coordinates with IoU-based losses: IoU, GIoU, DIoU and CIoU.
Precision, Recall and mAP
How detectors are scored: precision, recall and mean Average Precision over classes and IoU thresholds.
YOLO Tasks: Tracking, Segmentation and Pose
The same YOLO API does detection, instance segmentation, pose key points, oriented boxes and tracking.
Custom YOLO Training
Train YOLO on your own objects: collect images, label, split, train, evaluate.
YOLO Dataset Preparation and Annotation
YOLO format: one .txt per image with `class x_center y_center width height` normalised to 0–1.
YOLO Model Export and Deployment
Export to ONNX, TensorRT, OpenVINO, CoreML or TFLite and run on servers, PCs, phones or edge boards.
Spotted a mistake or want to add detail? Suggest an improvement — Content Contributors' edits are reviewed by editors before they go live.
Discussion (0)
Sign in or join free to ask questions and discuss this topic.