A novel and efficient model pruning method for deep convolutional neural networks by evaluating the direct and indirect effects of filters.