Although binary neural networks (BNNs) enjoy extreme compression ratios, there are significant accuracy gap compared with full-precision models. Previous works propose various strategies to reduce the ...