This article is published by NetEase Cloud .
From October 22nd to 29th, the world's top computer vision experts gathered in Venice to participate in the ICCV2017 International Computer Vision Conference to conduct intensive discussions on the latest achievements in the field. The conference proceedings also represent the latest development direction and the highest level in the field of computer vision. The two ICCV papers submitted by Di Xinhan, an image algorithm engineer of NetEase Cloud Security (Yidun), were included in the conference and were invited to participate in the seminar.
ICCV is the highest-level conference in the field of computer vision, hosted by IEEE and held every two years worldwide. For a long time, ICCV's paper acceptance rate is very low, and it is the highest recognized level among the three major international conferences on computer vision (the other two are CVPR and ECCV).
▲Di Xinhan at the ICCV scene
At the ICCV2017 "CEFRL: Compact and Efficient Feature Representation and Learning in Computer Vision" special report meeting, Di Xinhan attended and participated in the discussion of the development and achievements of cutting-edge compact and effective features. and semi-supervised features, etc. The special guests of the seminar include Yoshua Bengio, Professor Pascal Fua and other famous scientists in the deep learning AI field.
The core of Di Xinhan's two papers discusses technical methods to improve the accuracy of semi-supervised classification, and to improve the quality and diversity of generated images. This method is applied to the image recognition of advertisements, pornography, violence and terrorism, and politics-related images in NetEase Cloud Security (Yidun) "Content Security" product line. It further improves the accuracy and recall of image classification. At the same time, the GAN-Boost method is used to expand the number of difficult-to-collect images in the image training library and improve the image quality, reducing the cost of database collection for products.
In the paper, combining information entropy and GAN network, two hypotheses are proposed to combat the input uncertainty of GAN network under the theoretical framework of information entropy. The GAN-Boost training method proves that the two proposed methods reduce the loss of general GAN network information and improve the efficiency of GAN network using training data set information, and through experimental comparisons, it is confirmed that it improves the performance of GAN network image semi-supervised classification. The accuracy rate enhances the quality and diversity of images after GAN network training on small datasets.
NetEase Cloud Security (Edun) is the first in the industry to launch the third-generation smart content security product, and is currently leading the industry in the field of inappropriate content identification such as intelligent pornography detection, advertisement filtering, violence terrorism and political-related identification. NetEase Cloud Security continuously invests R&D resources to improve the accuracy of intelligent identification, and has been widely used in well-known products such as NetEase Cloud Music, Yizhuan, and OPPO App Store.
Learn about NetEase Cloud:
NetEase Cloud Official Website: https://www.163yun.com/
New User Gift Package: https://www.163yun.com/gift
NetEase Cloud Community: https://sq.163yun.com/