Oneself want to tensorflow based version of the SSD source modify one of the main network, will be replaced VGG mobileNetv3 formed mobilenet SSD, want to consult everybody a great god, and how to assemble? Is only need the VGG network module in the program replace mobileNetv3 module line, or need to change some other part of the

When I just replaced the VGG network module in the program mobileNetv3 module after the modification to it after fine - tuning training, training 30000 times ran out of the result is not very ideal, now paste some pictures of my operation, to show you a great god, hope to get guidance!



Figure 1 is in program I replace

Figure 2 is a type I download mobileNetv3 training

Figure 3 is I'm ready for training to fill in some of the parameters of the

Figure 4 is training after the completion of the test I get a mAP

CodePudding user response:

Wish you a great god can bring the help to solve this problem, I don't own tnesorflow API