Researchers and programmers across the globe are trying to improve the computer vision and make it as accurate as a human eye. Now, there is a new record set by the system of Microsoft Researchers in the ImageNet Challenge. The team published a paper in this week, which mentions how their computer vision system, based on the Convolutional Neural Networks (CNNs), for the first time overdid the abilities of humans to classify objects that were defined in the ImageNet challenge.
What was the ImageNet Challenge
The ImageNet Challenge is the challenge of computer vision, where systems have to do image recognition with minimal error. The images are to be identified on the basis of their category as well as their resolution and scale. The images might be under various layers. The computer vision system has to tackle with all such issues and still identify the image using certain algorithm. An example of such identification is given below.
The ImageNet challenge’s new winner is the team of Microsoft researchers in Beijing. The team was successful in achieving an error rate of 4.94% on the 1000-class ImageNet 2012 classification dataset. The dataset consists of 1.2 million training images, 50,000 validation images, and 100,000 test images. In the previous similar experiments, the error rate achieved by humans was estimated as 5.1%. This way, Microsoft researchers’ team was able to break the record of the human’s ability to identify images, by achieving a lesser error rate.
The paper published by the Microsoft researchers’ team is titled as “Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification”.
The team says about their research,
“To our knowledge, our result is the first to surpass human-level performance…on this visual recognition challenge…In this paper, we investigate neural networks from two aspects particularly driven by the rectifiers…..”
The race for developing the best algorithm for image recognition in computer vision continues even after this new record. Read more about ImageNet Challenge and what the Microsoft researchers team has to say about their algorithm in this blog.
