Abstract: We propose a universal building block of Convolutional Neural Network (ConvNet) to improve the performance without any inference-time costs. The block is named Diverse Branch Block (DBB), ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results