Since the development of open source, more and more developers have been sharing their projects and code proudly as well as free code. Users get the results of the project freely, contributors find achievements and values, and then more developers join in the use and joint development, which is a positive cycle, thus promoting the healthy development of the community ecology…

What to watch for this week:

  • The AgentMaker team just released PaTTA, which can easily improve test set scores by inserting a single line of code.
  • Quest2020 deployment complete version of the source code on GitHub open source, help AI fruit quality inspection.
  • DodgeFace is innovating again with more gameplay and hidden NPCS waiting for you

For a more detailed introduction and GitHub address link, please check it out.

PaTTA

^Cifar100 performs with TTA on Resnet18

PaTTA is a library of test-time Augmentation, supported by the fly-oar core framework, that is currently available in the areas of image classification, image segmentation, and feature point extraction. Its principle is to enhance the test data in different forms, and then take the average (or proportional fusion) of the results as the final result, so it can generally improve the stability and accuracy of the results.

Project Maintainer: AgentMaker Organization member (Cao Zhihao, Xiao Peikai, Wang Zirui)

Main Frame/tool component: PaddlePaddle core frame

Making links:

Github.com/AgentMaker/…

Lemon Classification Hiroshima, Japan

Quest2020 is deployed in combat

Fruit quality inspection is one of the key steps of fruit commercialization processing, and its inspection results will directly affect the effect and benefit of later transportation, storage and sales. Traditional quality inspection methods are mainly manual inspection and conventional machine learning algorithm. However, with the rapid development of artificial intelligence, quality inspection with deep learning ability can provide higher profits for enterprises.

Project Maintainer: Yu Hang (Grey Sauce)

Main frame/tool components: PaddlePaddle Core Frame, PaddleLite

Making links:

Github.com/hang2451412…

DodgeFace-EXQver

After last week’s PaddleHub, PaddleHub developer Technology expert (PPDE) Han Lei has created a new PaddleHub! Dodgeface-exqver can be understood as an upgrade to DodgeFace, in which players take on the role of the “Brother of the Two Bridges” in an attempt to avoid being tracked down by “Sir Tan” Police officer Tan. In this game, in addition to avoiding “Tan Sir”, players can also interact with the other two mysterious NPC high frequency, of course, this will not affect the outcome of “Tan Sir” caught, after all, driving norms to stay in the heart.

Project Maintainer: Han Lei (Ninetailskim)

Main framework/tool components: PaddleHub, OpenCV

Making links:

Github.com/ninetailski…

Review of selected projects in the last issue

PAPC (Deep Learning 3D Point Cloud Library)

PAPC, named after PA in Paddle and PC in PointCloud, is a 3D PointCloud processing kit built using a core framework by two undergraduate Paddle developers. As one of the recent new open source projects, 13 common point cloud models are built in a short time, perfectly covering the three visual tasks of 3D classification, segmentation and target detection of deep learning.

Project Maintainer: AgentMaker Organization member (Cao Zhihao, Xiao Peikai, Wang Zirui)

Main Frame/tool component: PaddlePaddle core frame

Making links:

Github.com/AgentMaker/…

ExtractLine (Image to Line drawing tool)

“The line draft can make the empty paper produce positive and negative shape, and can naturally outline the shape, god, light, color, volume and texture of the image with the length, width, depth and light, and proper tension. Painters of different attainments can control different pictures. The difficulty is unfathomable, and the variety is amazing.”

ExtractLine is such an open source project that can convert images into line drawings, so that the deep learning model of different parameters can control different pictures, the difficulty of small at your fingertips, and the change is also magical.

Project Maintainer: Peng Zhaoshuai (7 years term), Zheng Bopei (Mr. Zheng)

Main framework/tool component: PaddleHub

Making links:

Github.com/1084667371/…

Open source is not easy, Star encourages

If you like the open source projects recommended in this issue, you are welcome to give yourself the project point Star (Github link is in the corresponding project introduction), your support is also the motivation of the author to stick to open source sharing!

We will randomly select one partner from the new Star list of 5 projects in this week and give a set of t-shit + geek socks as a gift. The result of the lucky draw will be publicized in the project recommendation article next week.

Open source projects don’t have to be big or small. Big open source projects become industry leaders, providing comprehensive solutions and approaches, but small projects are just as exciting and creative.

If you would like to recommend your own project, simply introduce the project, send an email to the Paddler Community staff ([email protected]) for a chance to participate in the project recommendation, and your code can bring efficiency and value to others!

PaddlePaddle, based on baidu’s years of deep learning technology research and business application, is China’s first open source, leading technology and fully functional industry-level deep learning platform, including open source platform and Enterprise version. Feioar open source platform includes core framework, basic model library, end-to-end development suite and tool components, continuously open source core capabilities, providing a foundation for industry, academic, scientific research innovation. Flying Oar Enterprise edition is based on flying oar open source platform and has enhanced corresponding features for enterprise requirements, including EasyDL, a zero-threshold AI development platform, and BML, a full-function AI development platform. EasyDL is mainly for small and medium-sized enterprises. It provides a convenient and efficient development platform with zero threshold, abundant network and models preset. BML is a comprehensive, flexible, customizable and deeply integrated development platform for large enterprises.