yifan123 / ic-conv Goto Github PK

May I use other networks, such as Unet, not resnet, how should I use Ic_con and how to generate a pattern?

Hi
when i use ICConv to replace the traditional convolution has this problem:

self.conv = ICConv2d(pattern[pattern_index], inplanes, planes,kernel_size=kernel_size, stride=stride, bias=False)
TypeError: 'NoneType' object is not subscriptable

how can i deal with this problem

ATT

How do I define this "pattern_dist" ,if I only use ICConv2d module.Thanks.

ModuleNotFoundError No module named 'linklink'

Thanks for your great repo.
The web of pretrained weights is 404. I can't download the all pretrained checkpoints. Is there something wrong?
Looking for your reply.

Dr. Liu, I am reading your Inception Convolution with Efficient Dilation Search, the content is very well written, but about this convolution in the pattern parameters how to find the minimum representation error E, I have been puzzled, see the derivation of the last step in the formula with the L1 error of both weights Minimization, how is the weight obtained after that expansion, by training or some other way.
Thank you very much if I can get your reply!