From my understanding, the conv layer without gaussian_kernel will just have random weights.
For the conv layer with gaussian_kernel, the weights.data of conv are just covered by
gaussian_kernel values which give us better edge detection.
Github URL https://github.com/pairlab/CBS
torch.arange()
https://yongqiang.blog.youkuaiyun.com/article/details/106745713
a.repeat(repeats)
repeats: int或int数组 每个元素的重复次数。
https://www.cnblogs.com/nxf-rabbit75/p/10055932.html
a.view()
https://blog.youkuaiyun.com/york1996/article/details/81949843
.t()