Many detection methods such as Faster-RCNN and YOLO, perform badly in small objects detec-tion. With some considerable improvements in the originalframework of YOLOv2, our proposed SO-YOLO can solvethis problem perfectly.

As a convolutional neural network, SO-YOLO outperforms state-of-the-art detection methods both in accuracy and speed.

SO-YOLO performs well in detecting small objects compared with other methods.