Image recognition, deep learning, artificial intelligence has become a first-line big companies to win the strategic location

Absrtact: With the advent of the reading age, a large number of content generated in the pictures and video, the image recognition, understanding and analysis needs more and more intense. For a time, image recognition, in-depth learning, artificial intelligence has become a first-line large companies to take the strategic location, Baidu

With the advent of the reading age, a large number of content generated in the picture and video, the image recognition, understanding and analysis of the need for more and more intense. For a time, image recognition, in-depth learning, artificial intelligence has become a first-line large companies to win the strategic location, Baidu has in-depth study institute, Ali also "big Data" and "top of the data" Xu Zixuan dug to be vice President layout Data mining. For the general company, they often do not have artificial intelligence engineering and technical capabilities, but also to deal with a large number of picture data needs, this contradiction how to break?

As a micro-letter founding team member of the "Li Mingqiang", its 2B business "image recognition open platform" is to help enterprises to customize their own image recognition engine, enterprises need not employ artificial intelligence identification experts to build research and development team, the Web interface can operate. Li Mingqiang told 36 Krypton, "the advantage of this image to identify the cloud platform is that it lowers the threshold for the use of artificial intelligence, and can only be used by simple training with Excel non-technical people." ”

The open platform is based on the depth learning technology and engineering experience accumulated by the graph technology, including tagging system, multi-GPU parallel scheduling system, statistic and effect analysis platform, test platform and API, etc. Its specific operating procedures are: Enterprise users in the application account and the establishment of a new identification project, and then upload data (web or SFTP)-labeled data (their own label)-training model-through the cloud service call model. In order to ensure the effect, upload data and annotation data can be more than a few cycles, continuous optimization. Examples of uploading image data and tagging are as follows:

Pornography/non-pornography
Advertising, garbage/Normal
Scenery/people/food/car/digital/...
Watermark/No watermark
Beauty/Ugly Girl
European and American fan/Japanese-Korean fan/art fan/Chao Nu fan
Even the waveform of the stock: Head/Shoulder Bottom/arc bottom w/...
Wait a minute

Li Mingqiang This "image recognition open platform" to a smart child, enterprise users just upload a variety of applications in the image of the data to "teach him" (data training), the child to see more to learn better, the more reliable work. Tupou is responsible for hardware and software problems, algorithms and engineering problems.

The current technology partners have thunder, petal nets and shoot clouds. Take the thunder of the yellow picture monitoring, if using the traditional method, thunder will usually use dozens of of the yellow division, each day to review the yellow Division 20,000 pictures, and Tupou technology for the Thunderbolt to save 90% of the audit manpower, a TAI yellow machine can audit 50 million pictures a day, its single video processing time is less than 1 seconds, 100 photo processing time is less than 1 seconds. Add total up, the current daily processing volume of the map is 120 million pictures, 300,000 video, and relative to the manual audit 98% accuracy, in this processing speed, the accuracy of the picture is 99.5%.

The cloud service model provided by the technology is based on the number of calls, the higher the unit price at the end of the month. If the enterprise's daily call amount is 10,000, each charge standard is 1 minutes; The daily call amount to tens, 100 of the charge is 1 cents.

Founder Li Mingqiang is the original Tencent project manager and T4 technical expert. 7 years experience in Internet product project management, team management, product design and operation promotion; Tencent is a practitioner of agile research and development process, and has 3 years of experience in telecom industry, and joined Tencent's Guangzhou Research and Development Center in 05, and is also a founding member of micro-letter project.