Binarizing by Classification: Is soft function really necessary?

Binary neural network leverages the $Sign$ function to binarize real values, and its non-derivative property inevitably brings huge gradient errors during backpropagation. Although many hand-designed soft functions have been proposed to approximate gradients, their mechanism is not clear and there are still huge performance gaps between binary models and their full-precision counterparts. To address this, we propose to tackle network binarization as a binary classification problem and use a multi-layer perceptron (MLP) as the classifier. The MLP-based classifier can fit any continuous function theoretically and is adaptively learned to binarize networks and backpropagate gradients without any specific soft function. With this view, we further prove experimentally that even a simple linear function can outperform previous complex soft functions. Extensive experiments demonstrate that the proposed method yields surprising performance both in image classification and human pose estimation tasks. Specifically, we achieve 65.7% top-1 accuracy of ResNet-34 on ImageNet dataset, with an absolute improvement of 2.8%. When evaluating on the challenging Microsoft COCO keypoint dataset, the proposed method enables binary networks to achieve a mAP of 60.6 for the first time, on par with some full-precision methods.