A Novel Low-Bit Quantization Strategy for Compressing Deep Neural Networks

Table 3

Accuracy comparison on CIFAR100.

avb	BNN	XNOR	Ours
ResNet-34	48.81/78.32	53.28/81.29	61.33/87.22
ResNet-50	52.07/81.60	59.20/85.32	62.92/88.65