Jupyter notebook examples on image classification with quantized neural networks. The intent here is to give a better understanding of what kind of computations takes place when performing inference with a quantized neural network.
So far, the following notebooks are available:
- Basics for a gentle warmup
- Binarized, fully-connected MNIST for a deep dive inside a binarized fully-connected network
- Binary weights, 2-bit activations, fully-connected MNIST for demonstrating what happens when we go to 2-bit activations
- Binarized, convolutional GTSRB for an introduction to convolutional and pooling layers
- Mixed 1-bit weights/2-bit activations and 8-bit weights/activations ImageNet for a quantized AlexNet on the ImageNet dataset.