Academia Sinica Logo

Groundbreaking results in artificial intelligence achieved by international team with Taiwanese researchers! AS Distinguished Research Fellow Mark Liao and Postdoctoral Scholar Chien-Yao Wang from the Institute of Information Science worked with Alexey Bochkovskiy from Russia to develop YOLOv4, currently the fastest and most accurate object detection algorithm. YOLOv4 has an average precision (AP) rate of 43.5 percent, 10 percent higher than the previous version (YOLOv3).

Abbreviated from “You Only Look Once”, YOLO has been highly favored by developers since its first version was released in 2015 because it allows computers to detect objects’ type and location after just one glance at the picture or image.

Chien-Yao Wang started improving YOLOv3 in 2019. Using a machine learning model approach, he optimized the propagation path to reduce computation and increase frame rate plus detection accuracy.

Dr. Wang’s results caught the attention of Alexey Bochkovskiy, who built and maintained the YOLO website. The two of them started working together in November 2019, and announced the completion of YOLOv4 this past April.


The results of Microsoft COCO Object Detection tasks show that YOLOv4 is highly improved and can identify objects both faster and with higher accuracy.

Since the source code was shared on an open-source platform, tens of thousands of people around the world have tested YOLOv4 and developed their systems and products. According to Mark Liao, YOLOv4 has been used in the development of “Smart City Traffic Flow Solutions”, a collaborative project with ELAN Microelectronics Corporation to enhance smart city innovation in Taiwan.

The team’s research paper on this discovery was published in April, and can be found at the following website: https://arxiv.org/abs/2004.10934