your repo is very useful for us to use this model,because c++ is more efficient than python .i am appreciated your work ,it will help me a lot.you say ' The python script for converting the trained model to the frozen graph will be provided. 'and 'A pretrained frozen graph will be provided based on the original code in repo.' so when can we see the released code and model .and you said you have finished a nonmax script,so could you upload to your repo.