Distribution_Loss

Source code for paper "Regularizing Activation Distribution for Training Binarized Deep Networks"

Code modified from the code for the original BNN paper.

Train BNN-DL for Alexnet ImageNet

CUDA_VISIBLE_DEVICES=0 python custom_main_binary_imagenet.py --seed 1 --model alexnet_binary_vs_xnor --batch-size 256 --batch_size_test 100 --infl_ratio 1 --distrloss 2. --distr_epoch 60 --epochs 65 --gpus 0

You can download checkpoint from this link. The top-1 accuracy is 47.9%, and top-5 accuracy is 71.9%.