Yolov3 trains its own data set

yolov3 Is in yolov1 And yolov2 Based on the modified end-to-end Target detection algorithm .

yolov3 Use Darknet53 As a backbone feature extraction network , The network has the following two characteristics ：
1、Darknet53 Residual networks are used Residual. The residual network is first convoluted with a kernel size of 3X3、 In steps of 2 Convolution of , The convolution compresses the width and height of the input feature layer . At this point, you will get a feature layer , This feature layer is usually named layer. Then, the characteristic layer will be tested again 1X1 Of convolution and once 3X3 Convolution of , And add this result to layer, At this time, the residual structure is formed . Through constant 1X1 Convolution sum 3X3 Convolution and superposition of residual edges , Can greatly deepen the network . The characteristic of residual network is easy to optimize , And it can improve the accuracy by increasing the depth . Its internal residual block uses jump connection , It alleviates the problem of gradient disappearance caused by increasing depth in depth neural network .

2、Darknet53 Each convolution part of the uses a unique DarknetConv2D structure , Every convolution is done l2 Regularization , After the completion of convolution BatchNormalization Standardization and LeakyReLU. ordinary ReLU Is to set all negative values to zero ,Leaky ReLU It gives all negative values a non-zero slope . We can express it mathematically as ：

This topic focuses on operation , Next, we will talk about how to use this yolov3 Algorithm to achieve their own data set training and detection .

Source code address ：

GitHub - ultr