X-dimensional pocket
It is the mission of seven cows to make the data scene deep and really help users shorten the distance from idea to product. From 2011 to now, we have built a scenario PaaS around data management, and we have served more than 500,000 customers, carrying more than 200 billion images and 1 billion hours of video.
for Internet developers, one-stop online data management must include 0 operations, high-availability, and high-performance data processing services for large volumes, so that you can easily respond to real-time, asynchronous processing scenarios for images, audio and video, and other types of data. At present, the data processing services provided by seven cattle have been processed nearly times, in addition to providing basic data processing, but also based on container technology to create a scalable, easy to deploy, high degree of freedom of the custom data processing access platform, and seven of cattle data processing services seamlessly compatible with the use of a rich media ecosystem around data development.
"x-dimensional pocket" is the introduction of these rich media data processing services of a column.
In recent years, mobile social class, UGC-type platform, the number of apps emerge, but the face of a large number of daily users upload pictures, traditional Master Huang still busy come? Creating a healthy social ecosystem has become a key to the long-term development of these platforms and apps, and it is an urgent task to effectively and accurately identify and eliminate obscene and pornographic information.
Seven Cow cloud and focus on the image recognition technology Tupou technology launched a picture Kam Yellow service, can be judged to save in the seven Qiniu pictures are pornographic, sexy or normal, and give a judgment of the score, the higher the score is judged as the probability of the category is greater.
How is the image yellow?
image Yellow is a kind of image recognition technology based on deep learning. Simply put: imitate the human brain mechanism to analyze the data, so that the machine learns to extract the characteristics of discriminant, thereby eliminating the process of manual selection.
The working principle can be summed up as: -Image Image analysis (skin tone, posture, scene); -Interactive behavior analysis (file name, open/upload time); -Feature library analysis (image fingerprinting with different cubby pairs); -operation and maintenance Management (user report, administrator identification, identification of the yellow division); -Evolution (update the Trust/pornography site list, update the image truststore, update the file name keyword blacklist).
the ability and advantage of Photo yellow service
1. Cost reduction at the same time, efficiency greatly improved.
Kam Huang Average daily audit processing of pictures only in 30,000 or so, the machine can be examined 900 million a day. Yellow service based on large-scale GPU over-counting platform, batch large concurrent up to 0.02 seconds/Zhang response speed, while ensuring stability, greatly improve the audit efficiency, effectively reduce enterprise operating costs.
2. Accuracy is increased, the number of false positives is low. According to the results of the third-party service provider evaluation, the seven-ox platform has more than 99.5% accuracy, can replace more than 90% of human audit, and has a strong self-learning ability, through continuous learning to improve the accuracy of identification.
3. Full-time work.
due to the particularity of the work, artificial Kam Huang is difficult to work for a long time, and machine Kam Yellow can do 7 * 24 hours of sleepless.
Application Scenarios
UGC Platform Image review
social applications, vertical communities, and other UGC-based platforms need to review the content of images uploaded by users. For the general platform, the yellow-related image only accounts for 1% or less of the total content, but requires one by one audits of all content to eliminate risk and a huge human consumption. By accessing the yellow service, the machine classifies the image content as pornographic/sexy/normal, and gives the machine determination and the machine is unable to determine the part, the auditor only need to the machine difficult to determine the part (about the total amount of 1%~5%) audit. In this way, the machine is replaced by the majority of audit manpower, ensuring reliable and stable audit work and reduce labor costs by more than 90%.
open seven ox kam Yellow three step walk On The seven cattle platform to turn on yellow service, resource storage and resource processing seamless, to achieve zero programming yellow, no need to go through lengthy registration, verification process.
First step: Click to start using
Step Two: Get the picture URL
Step Three: Add command after the image URL nrop
description of the return value:1) The Label value indicates the machine's recognition of the image. Label=0, representing the picture as a pornographic image; Label=1, representing the picture as a sexy picture; label=2, which represents the picture as normal. 2) The rate value indicates the likelihood that the picture conforms to the machine's recognition results. Example: label=0,rate=0.9987, which indicates that there are 99.87% of possible pornographic images. 3) The Review value indicates whether the picture/video needs to be reviewed manually. Review=true, on behalf of the machine to the image of the judgment of less than 99.5%, the proposed review, Review=false, said the machine to the image of the judgment of more than 99.5%, higher than the artificial, is not reviewed .
More details Click to view:
https://portal.qiniu.com/create
A large sword---x-dimensional pocket in the cloud service to eradicate the small yellow figure
