YOLO v3 uses a more complex network architecture and incorporates skip connections, which improves the accuracy of the predictions -
• Multi-scale predictions: YOLO v3 uses anchor boxes of different sizes to detect objects at different scales, which improves the accuracy of the predictions
• Improved non-maximum suppression: YOLO v3 uses a more sophisticated non-maximum suppression algorithm, which helps to eliminate duplicate detections
• Improved training: YOLO v3 uses a technique called label smoothing to improve the generalization of the network and reduce overfitting