This project is a collection of projects I have run in a while.Including the mainstream direction of vision, such as image classification, target detection, etc.

I would like to thank one of my partners yuyufei88for providing the Android program.

Currently, only two frameworks, Mxnet and Ncnn, are used. The project structure is as follows：

• Mxnet/resnet18
• ncnn
• Android_demo

Here are yolov5 of pytorch, naodet of pytorch, yolov3_mobilenet of mxnet, and their ncnn versio. The project structure is as follows：

• naodet
• naodet_ncnn
• yolov2ncnn
• yolov3_mobilnet
• yolov3_mobilnet_ncnn
• yolov5
• yolov5_ncnn
• Android_demo

yolov2ncnn is a detailed process of yolov5 to ncnn. There is a process of writing yolov3 to ncnn in yolov3_mobilnet.

There are AttentionOCR suitable for scene text recognition, general text recognition, and EasyOCR that can recognize multiple languages. The project structure is as follows：

• AttentionOCR
• AttentionOCR_flask
• EasyOCR
• PaddleOCR

These are two very good target tracking projects. Real-time running on the server side is not a problem, and it is probably maintained at about 30fps. The project structure is as follows：

• FairMOT
• Towards_Realtime_MOT

These are two very good papers, in which TEA personally sorted them out, and the experimental results are credible.

• ECO-caffe
• TEA

Here is yolov5 for detection, using arcFace for recognition, and Insight Face.

• yolov5_arcFace
• yolov5_face
• Insight Face

In yolov5_face, face detection is performed after yolov5 is converted to ncnn.

This library will be updated in the future to make it richer and more complete. The collation of papers will be updated in the WeChat public account.

All articles and code parts that involve others have links to the original text. If there is any forgotten part, please point it out and I will improve it.