Conditional convolution (Dynamic convolution) in tensorflow2.2.0. This depository implements the method described in the paper:
CondConv: Conditionally Parameterized Convolutions for Efficient Inference
Brandon Yang, Gabriel Bender, Quoc V.Le, Jiquan Ngiam
Source PDF
Meanwhile, the softmax with a large temperature for kernel attention introduced by Dynamic Convolution: Attention Over Convolution Kernels is adopted.
Another similar paper: DyNet: Dynamic Convolution for Accelerating Convolutional Neural Networks.
You can start according to the default arguments by python main.py. Or specify the arguments:
pythonmain.py--archcond_cifar_resnet--num_layers56--num_experts3--datasetcifar10--num_classes10