yolov3 and yolov5
These are ecosystem siblings representing successive generations of the same architecture—YOLOv5 is the newer iteration that supersedes YOLOv3, both maintained by Ultralytics with overlapping export capabilities (PyTorch to ONNX/CoreML/TFLite) but YOLOv5 offering improved accuracy and speed that make it the preferred choice for most new projects.
About yolov3
ultralytics/yolov3
YOLOv3 in PyTorch > ONNX > CoreML > TFLite
Implements multi-scale feature pyramids and anchor-based detection with spatial attention mechanisms for improved accuracy on small objects. Provides built-in training pipelines with data augmentation, mixed-precision support, and batch normalization optimization across PyTorch, ONNX, CoreML, and TFLite export targets for seamless deployment across cloud and edge devices.
About yolov5
ultralytics/yolov5
YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite
Supports object detection, instance segmentation, and image classification tasks across diverse hardware through unified training and inference pipelines. Built on PyTorch with automated model optimization for deployment via PyTorch Hub, offering pre-trained weights and reproducible training from scratch with configurable architectures (Nano to Extra-Large). Integrates native inference acceleration through TorchScript, Docker containers, and cloud platforms like Google Colab and Kaggle for streamlined development and deployment workflows.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work