Can't "coder.loadDeepLearningNetwork" import a custom network model?

Question

cui,xingxing 2019-10-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/485234-can-t-coder-loaddeeplearningnetwork-import-a-custom-network-model

回答： David Fink 2019-10-14

When I am going to use GPU Coder to generate cuda code for my custom depth model, matlab2019b unexpectedly terminates, and matlab2019a reports that fopen generates invalid identifier), how can I solve it? In addition, the custom model I imported is also very simple, bvlc's googlenet, not comes with pre-trained googlenet, my code is as follows, the model see attachment (can be loaded to matlab normally, each layer also supports code generation)

my gpu environment is OK:

envCfg = coder.gpuEnvConfig('host');
envCfg.DeepLibTarget = 'cudnn';
envCfg.DeepCodegen = 1;
envCfg.Quiet = 1;
coder.checkGpuInstall(envCfg);

and the code deploy is:

function out = googlenet_predict(in) %#codegen
persistent mynet;
if isempty(mynet)
    mynet = coder.loadDeepLearningNetwork('bvlc_googlenet.mat','googlenet'); % can directly load to matlab workspace,but can't load to GPU Coder to generate??? 
end
% pass in input   
out = predict(mynet,in); 

The code that generates the configuration is：

cfg = coder.gpuConfig('dll');
cfg.TargetLang = 'C++';
cfg.DeepLearningConfig = coder.DeepLearningConfig('cudnn');
cfg.DeepLearningConfig.AutoTuning = true;
codegen -config cfg googlenet_predict -args {ones(224,224,3)} -report