QRCode NCNN

WeChat-QRCode-NCNN

WeChatQRCode is a QR code engine launched by the WeChat AI team, but the original work uses caffe and is open sourced from OpenCV.

(!!! This original engine was not made by me, I just transplanted it!!!)

I am a fan of ncnn, and I couldn’t bear to see Tencent’s products being copied by caffe ntr, so I put them on ncnn to get to the bottom of things.

Public account introduction: Portal

Original Code: Portal

The work of this project: Move the reasoning part of the engine from caffe to Tencent's own ncnn

Brief introduction

A brief introduction to the workflow of this version of the engine:

1. QR code detection, ncnn runs the detect model
2. QR code super score, ncnn runs SR model
3. QR code decoding, using zxing-cpp

advantage

1. The call is simple:

cv:: Ptr <cv::wechat_qrcode::WeChatQRCode> detector = cv::makePtr<cv::wechat_qrcode::WeChatQRCode>( " assert/detect.param " , " assert/detect.bin " , " assert/sr.param " , " assert/sr.bin " );
vector<cv::Mat> points;
vector<string> res = detector-> detectAndDecode (img, points);

2. The project is small: the detect model + sr model add up to only 960KB, which is very small even if it is directly compiled into the project.

Effect

Note: The mobile phone is Redmi 7, Snapdragon 632 (a super typical 100-yuan phone). The code is detected and decoded in real time, so it may feel laggy when used. If the equipment is good, it should not be laggy, and in actual use, it is One-shot, no one should be running real-time.

The APP has exported a debug version, here

Work content

• x86 platform work
• android platform work

statement

1. The original inference engine belongs to Tencent WeChat AI team's open source on OpenCV
2. What I did was just change the reasoning engine from caffe to ncnn.
3. For commercial use, please refer to the official License! ! !
4. Open source is not easy (please beg for star)