As to why programmers like to keep cats, netizens give many answers, such as sucking cats while writing code, which is more efficient; Owning a cat prevents hair loss; Easier to get a girlfriend…

Of course, in addition to like raising cats, programmers also like to open imagination, play their own expertise, to the “cat owners” arrangement on all kinds of high-tech — automatic feeding machine, automatic eating machine, automatic massage instrument…… They even tested cats while training models to identify their emotions and understand their language.

There’s an AI app out there that was originally developed by a programmer who wanted to identify a cat’s pen for his master. The power behind it comes from the MegFlow streaming computing framework.

Megvii’s open source deep learning framework MegEngine has recently opened its MegFlow streaming computing framework to help AI algorithm developers quickly implement THEIR AI models.

MegFlow is a streaming computing framework for computer vision applications, providing a set of visual parsing services for rapid DEPLOYMENT of AI applications. AI application developers can customize features such as invoice scanning and open fire detection in as little as 15 minutes based on MegFlow’s image and video parsing services.

The conventional AI algorithm delivery process is generally divided into four steps: model training, SDK packaging, business integration and delivery acceptance. MegFlow, on the other hand, provides a more concise visual application implementation process. Users can directly build computation diagrams in Python without having to worry about C++ and graph optimization. This eliminates SDK packaging process and enables fast algorithm delivery.

MegFlow can provide effective solutions to frequent problems in AI algorithm engineering implementation, such as performance tuning, security, model encryption, etc., and effectively improve the engineering efficiency. MegFlow features security, reliability, ease of use, and rich semantic support, helping AI applications to land quickly.

Safe and reliable

In terms of technology selection, MegFlow’s R&D team investigated a variety of technical solutions, and finally chose Rust asynchronous ecology, which is safe and has zero extra cost, to ensure the safety and performance of MegFlow.

Simple and easy to use

MegFlow supports Python plugins, and developers only need to write synchronous Python programs to implement Python plugins that can be asynchronously scheduled by MegFlow. MegFlow also provides a set of visual debugging tools based on the Web UI to effectively improve the efficiency of model deployment.

Rich semantic support

In terms of expression capability, MegFlow supports static graphs, dynamic graphs and shared graphs, supplemented by universal plug-ins with functional semantics such as Demux, Reorder and Transform, providing rich semantic support for building diversified AI services.

MegFlow now has built-in AI applications out of the box, such as battery car detection and pet pen detection:

  • Battery car detection application provides intelligent management tools for property managers. If the camera detects battery car entering the elevator, the system will send a notice to remind management personnel and effectively eliminate fire safety hazards.
  • Pet pen detection currently supports cat detection, and an alarm will be sent when a registered cat leaves the pen.

More computer vision-related AI models and applications will follow. To learn more about MegFlow and how to use it: github.com/MegEngine/M…