Download package: https://www.cntk.ai/OnnxModels/mnist.tar.gz
Model size: 26 kB

This model predicts handwritten digits using a convolutional neural network (CNN).

The model has been trained on the popular MNIST dataset.

The model is trained in CNTK following the tutorial CNTK 103D: Convolutional Neural Network with MNIST. Note that the specific architecture used is the model with alternating convolution and max pooling layers (found under the "Solution" section at the end of the tutorial).

Input image of the shape (1x28x28)

Output is a (1x10) array

Resize the input image to a (1x28X28) array of type float32, scaled to between [0,1] (computed by: image/255).

Route the model output through a softmax function to map the aggregated activations across the network to probabilities across the 10 classes.

Sets of sample input and output files are provided in .npz format (test_data_*.npz). The input is a (1x28x28) numpy array of an MNIST test image, while the output is an array of length 10 corresponding to the output of evaluating the model on the sample input.

MIT