Blog

Training YOLOv8 Pose on a custom dataset allows you to detect objects and predict specific keypoints that describe their pose or structure. Instead of being limited to the standard human body keypoints used by pretrained pose models, you can define your own classes, keypoint locations, and skeleton structure for applications such as animal pose estimation, […]

YOLOv8 release notes and updates document the changes made to the Ultralytics package that supports YOLOv8 and other Ultralytics YOLO models. Since YOLOv8 was introduced on January 10, 2023, the surrounding Ultralytics software has received frequent updates covering training, prediction, export, deployment, bug fixes, compatibility, and performance. YOLOv8 remains supported inside the modern Ultralytics framework

The YOLOv8 CSPDarknet backbone is responsible for extracting meaningful visual features from input images before those features are passed to the neck and detection head. Its design builds on CSP-style feature extraction principles while using C2f modules to improve gradient flow, feature reuse, and computational efficiency. This backbone helps YOLOv8 identify useful patterns at different

Exporting YOLOv8 to ONNX format allows a trained Ultralytics model to run outside the normal PyTorch environment. ONNX, or Open Neural Network Exchange, provides a standardized model representation supported by runtimes and deployment tools such as ONNX Runtime, OpenCV DNN, TensorRT conversion pipelines, and other inference systems. Ultralytics provides built-in ONNX export through both the

YOLOv8 Non-Maximum Suppression, or NMS, is a post-processing step used to remove redundant overlapping bounding boxes from object detection predictions. A detector may initially produce several boxes around the same object. NMS compares their confidence scores and overlap, keeps the strongest detection, and suppresses weaker overlapping boxes. Current Ultralytics prediction settings expose parameters such as

The YOLOv8 IoU threshold controls how strongly overlapping bounding boxes are treated during prediction post-processing and validation. IoU, or Intersection over Union, measures the amount of overlap between two bounding boxes. In Ultralytics prediction mode, the iou argument is primarily used as the IoU threshold for Non-Maximum Suppression, where overlapping predictions are filtered to reduce

The YOLOv8 Task Aligned Assigner is a training component used to match model predictions with ground-truth objects. It helps determine which prediction points should be treated as positive samples during training by considering both classification confidence and localization quality. This improves the alignment between object classification and bounding box prediction, allowing YOLOv8 to learn more

The YOLOv8 SPPF module is an important part of the backbone that helps the model capture information from a wider spatial area without adding heavy computational cost. SPPF stands for Spatial Pyramid Pooling Fast. It uses repeated max-pooling operations and feature concatenation to collect multi-scale contextual information before passing the processed features to later stages

YOLOv8 loss functions determine how the model measures prediction errors during training and how strongly those errors influence weight updates. For standard YOLOv8 object detection, the main reported loss components are box loss, classification loss, and Distribution Focal Loss (DFL). Current Ultralytics detection loss code explicitly maintains these three components and combines them using configurable

The YOLOv8 decoupled head is a key component of the model’s detection architecture. It separates object classification from bounding box regression so each task can be optimized independently. This design helps YOLOv8 improve detection accuracy, localization quality, and training efficiency while maintaining fast inference performance. Introduction to the YOLOv8 Decoupled Head The detection head is