Shulou(Shulou.com)06/02 Report--

Xiaocha from Aofei Temple

qubit report| QbitAI

Today, there was a major earthquake in AI academia.

YOLO algorithm author Joseph Redmon announced on Twitter that he would stop all CV research because his open source algorithm has been used for military and privacy issues. This posed a great test to his morals.

YOLO is one of the most well-known object detection algorithms in computer vision and is widely adopted by the industry because of open source.

Author Joseph Redmon won the CVPR People's Choice Award in 2016 and the CVPR Best Paper Honorable Mention in 2017 for his algorithm.

YOLO and its improved algorithm have been widely cited in academic circles, and the total number of citations of Redmon's three related papers has exceeded 10,000.

The sudden withdrawal of such an influential scholar shocked the academic community.

Fast.ai founder Jeremy Howard expressed his feelings: "I've never seen anything like this before. "

A French tech company CTO paid tribute to Redmon's faith.

Redmon's flamboyant personality in the past may have set the stage for his decision today.

About YOLO's father

Jeseph Redmon graduated from Middlebury College with a major in computer science and a minor in mathematics. In 2013, he entered the University of Washington to study computer science for a master's degree and then a doctoral degree until 2019.

During this period, he and his mentor Ali Farhadi jointly proposed and improved YOLO algorithm.

His main research areas are object detection, image classification and model compression. Redmon's announcement to withdraw from CV research is basically equivalent to starting a new business in the future.

His resume looks like this:

Joseph Redmon's past internships also shine.

He interned at Google Brain, working on real-time robotic grip detection. He then moved on to AI2 (later XNOR.ai), where he proposed the binary approximation convolutional neural network XNOR-Net, which reduces the amount of computation required to classify images on mobile devices.

Redmon received a Google PhD Fellowship in 2018 for developing CV tools for practical applications better and faster, and was awarded a fellowship in the category of Machine Perception, Speech Technology, and Computer Vision.

Redmon himself has appeared on the TED stage to introduce CV algorithms for fast object recognition.

A Fast YOLO Algorithm

YOLO full name is You Only Look Once (you only need to look once), from the name can also see the advantage of this algorithm speed, so in many edge devices, YOLO algorithm is widely used.

Unlike Fast R-CNN, another well-known target detection algorithm, YOLO uses a "one-step" strategy to generate the category and location of the target object simultaneously.

YOLO has two advantages over Fast R-CNN:

Fast speed: detection rate of 45 frames per second, which can be used in real-time video detection, even up to 155 frames on smaller models;

Good versatility: networks trained on real image data can be used in fictional paintings.

However, YOLO also has certain limitations: the accuracy rate is not as good as Fast R-CNN, only one object can be detected in each square, and for objects with irregular edges, it will affect the recognition of surrounding objects.

Redmon later developed YOLO9000, YOLOv3 and other algorithms on the original YOLO technology, expanding the types of detected objects and improving the accuracy of the model.

Can you still use YOLO in the future?

Whether Redmon's sudden decision will affect the future use of YOLO. Don't worry about that for a while.

The author simply announced his withdrawal from CV research, and none of the open source projects on his personal GitHub page were closed. Even if the author does not maintain it in the future, given YOLO's huge influence, other developers in the open source community will certainly continue to maintain it in the future.

Redmon himself is unlikely to apply for a patent for YOLO algorithm, because Redmon is only expressing dissatisfaction with the scope of AI technology applications in this manner, and applying for a patent does not prevent military and unethical commercial use.

As AI develops to this day, moral problems are becoming more and more obvious.

San Francisco passed a law banning police from using Face Recognition technology in law enforcement, ignoring the condemnation of AI monitoring technology in the classroom and warning AI not to cross the line.

The withdrawal of YOLO's father is another wake-up call from academia about AI ethics.

- End-

https://www.toutiao.com/i6796131431694205451/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.