NVIDIA TensorRT Sample "sampleMNIST"

The sampleMNIST sample demonstrates how to:

• Perform the basic setup and initialization of TensorRT
• Import a trained Caffe MNIST digit classifier
• Perform simple input preprocessing before running the engine, like mean normalization in this case
• Verify the output from the inference

Usage

This sample can be run as:

./sample_mnist [-h] [--datadir=/path/to/data/dir/] [--useDLA=N]

SampleMNIST reads two Caffe files to build the network:

• mnist.prototxt - The prototxt file that contains the network design
• mnist.caffemodel - The model file which contains the trained weights for the network
• mnist_mean.binaryproto - The binaryproto file which contains the means

By default, the sample expects these files to be in data/samples/mnist/ or data/mnist/. The list of default directories can be changed by adding one or more paths with --datadir=/new/path/ as a command line argument.