Train a trained object detector with more epochs

Question

Hao Yu Wang 2019-2-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/445755-train-a-trained-object-detector-with-more-epochs

回答： Hao Yu Wang 2019-3-15

采纳的回答： Hao Yu Wang

在 MATLAB Online 中打开

Hello,

I already trained a object detector with following code:

% Training option settings
options = trainingOptions('sgdm', ...
    'MiniBatchSize', 1, ...
    'InitialLearnRate', 1e-3, ...
    'MaxEpochs', 5, ...
    'VerboseFrequency', 200, ...
    'CheckpointPath', tempdir);
detector = trainFasterRCNNObjectDetector(trainingData, layers, options)

I would like to train more epochs based on above codes and do not re-run the script.

How can I do?

Thank you!

Eric

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Hao Yu Wang 2019-3-15

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/445755-train-a-trained-object-detector-with-more-epochs#answer_365516

在 MATLAB Online 中打开

First of all, the title is not clear.

I found the answer which is called resume training. (https://www.mathworks.com/help/deeplearning/ug/resume-training-from-a-checkpoint-network.html)

But my condition is different from the offcical website example. I used function trainFasterRCNNObjectDetector and the type of my network is DAGNetwork. As a result, the solution for my situation is shown below

load('rcnn.mat');                 % Load previously trained network.
net = layerGraph(rcnn.Network);   % Transform DAGNetwork to make it excutable for trainFasterRCNNObjectDetector.
rcnn = trainFasterRCNNObjectDetector(trainingImages, layerGraph(net), options);
save('rcnn.mat', 'rcnn');         % Save trained network.