I used the tensorflow network guide to train the mnist model as here : https://movidius.github.io/ncsdk/tf_compile_guidance.html
I copied all the files from the training location(on a cluster node) to my computer, and tried to compile using :
mvNCCompile mnist_inference.meta -w mnist_inference -s 12 -in input -on output -o mnist_inference.graph
This gave me an error saying:
/usr/local/bin/ncsdk/Controllers/FileIO.py:52: UserWarning: You are using a large type. Consider reducing your data sizes for best performance
"Consider reducing your data sizes for best performance\033[0m")
I had no issues while building and using the examples in the workplace. I am using Ubuntu 16.04 for the work.
Should I reduce the dataset size used for training and then try deploying?
Another question, is if it can handle Inceptionet-v3 trained on the larger Imagenet data, why should it not be able to handle the MNIST dataset (I believe Imagenet dataset is larger, I might be wrong)? Was it not trained on the entire Imagenet data, but a subset?
@asingh42 The warning about a "large type being used" is an outdated and should be removed in the next release. The warning has nothing to do with the training dataset. The message was originally used to warn the user when a model contained 32 bit floating point types. Since the weights are converted to fp16, this warning is no longer valid and you can ignore this error.
@asingh42 You should be able to use the mvNCCompile tool to create a graph file which is can then use in your C or Python app. The default name of the file will be graph. MvNCCompile should create that file regardless of the warning unless there is another underlying issue.