Image recognition method and related device
A picture recognition and picture technology, applied in character and pattern recognition, image communication, instruments, etc., can solve the problems of low audit efficiency, time-consuming and labor-intensive, etc., to improve the efficiency of pornographic audit and reduce the possibility of bottlenecks , reduce the effect of identifying resources
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 2
[0078] Embodiment 2, a kind of identification method of illegal picture;
[0079] In practical applications, with the popularization of artificial intelligence deep learning, deep learning is now used to intelligently identify whether there is pornographic content in pictures. Usually deep learning consumes more computer resources, but for a live broadcast platform, there are tens of thousands of live broadcast rooms at the same time, and it is necessary to take screenshots of the content of the live broadcast room every preset time period of 30 seconds to check whether it contains pornographic content. Therefore, a dedicated deep learning machine is used, which has a large amount of graphics processing unit (GPU) computing resources. For the pornographic service framework, it is necessary to make good use of deep learning machines, and make the program more robust, and at the same time, it can cover all the live broadcast rooms of the live broadcast platform. Therefore, this...
Embodiment 3
[0121] Embodiment 3, a method for managing multiple identification machines
[0122] In addition, in practical applications, for the pornographic service framework, due to the large number of live broadcast rooms, and for each live broadcast room, a picture in the video of the live broadcast room will be intercepted every preset duration such as 30 seconds. To call the yellow identification service, then a machine that recognizes as a service cannot meet the online needs. Therefore, multiple GPU-based recognition machines are required (deep learning uses GPUs for a large number of calculations). For the identification service framework, a method needs to be designed to schedule different identification machines according to the request, and at the same time, it can also serve the purpose of disaster recovery and backup for multiple machines. Therefore, the embodiment of the present application also provides a method for managing multiple identification machines, and the speci...
Embodiment 4
[0175] Embodiment 4. A task scheduling method;
[0176] In addition, for the pornographic service framework, when the number of simultaneous broadcasts in the live broadcast room is relatively large, it will cause a lot of requests, and it is very likely that the number of requests at the same time is very large, which will cause many requests to time out and not be processed. Then, based on the machines with multiple identification services, how to design reasonably so that all the request tasks will not cause timeout or discard, and at the same time, the identification tasks can be performed at full capacity for multiple identification machines without causing a single identification There are a lot of tasks while other machines are idle. Based on this, this paper designs all yellow identification and identification requests in the form of cache queues on the master machine, and also designs load balancing scheduling control on the master machine, so that the tasks are d...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More