Proof of concept XNOR-Net demo on MNIST dataset.
Download the mnist dataset and unpack it to path_data.
- mnist_cnn: ordinary convolutional network
- mnist_bwn: binary weight network
- mnist_xnor: xnor-net
- torch version
cd torch/xnornet_cpu/
th main.lua -dataDir <path_data> -modelName <mnist_xnor>- mxnet version
cd mxnet/
python main.py --network='mnist_cnn'- torch version
| Model type | Testing accuracy |
|---|---|
| mnist_cnn | 98.81% |
| mnist_bwn | 98.38% |
| mnist_xnor | 96.64% |
- mxnet version
| Model type | Testing accuracy |
|---|---|
| mnist_cnn | 99.02% |
| mnist_bwn | 98.92% |
| mnist_xnor | 98.26% |