Abstract: We propose reparameterized refocusing convolution (RefConv) as a replacement for regular convolutional layers, which is a plug-and-play module to improve the performance without any ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results