Fork自 PaddlePaddle / PaddleDetection
Add the use_nnpack parameter in ExpandConvLayer, so that the convolution calculation can be switched to the NNPACK function.