Pytorch-CapsuleNet

Pytorch-CapsuleNet copied to clipboard

lmxhappy

changed the matrix multiplication in line 60 for better performance

1

Instead of creating the matrix W, a large matrix, I have calculated the matrix multiplication in an iterative way. So for batch size 100, the memory allocation has reduced from...

Z4HRA-S

How to change the number of categories of a classification to apply to your own dataset?

3

File "C:\Users\23671\Desktop\CNN\self_conv\DDR\CNN_2d.py", line 405, in forward u = u.view(x.size(0), self.num_routes, -1) RuntimeError: shape '[2, 1152, -1]' is invalid for input of size 1125888 When applied to my own dataset, the...

jiangye-git

After a few Epochs, Loss Would Be NaN

3

Hello, I tried to test your code but after a few epochs (8 at most) the loss of the network turns out to be NaN. Has anyone else faced this...

AzadeAlizade

胶囊网络前N-1次迭代停止梯度的问题

1

目前看到的多个版本胶囊网络都是在前N-1次迭代时停止u_hat的梯度，只在最后一次开放梯度。只有作者这个版本没有这样操作，请问作者是否清楚这里的原因

sodleave

Why need to get the mean value of the batch dimension?

1

``` b_ij = b_ij + a_ij.squeeze(4).mean(dim=0, keepdim=True) ```

xianxuhua

https://github.com/jindongwang/Pytorch-CapsuleNet/blob/93996119a224e9c3d274df88735f1acb36592d97/capsnet.py#L79 To make it stable, following is suggested: ```python def squash(self, input_tensor, epsilon=1e-7): squared_norm = (input_tensor ** 2 + epsilon).sum(-1, keepdim=True) output_tensor = squared_norm * input_tensor / ((1. + squared_norm)...

saswatpp

In file capsnet.py => function squash, ``` def squash(self, input_tensor): print(f'input_tensor.shape: {input_tensor.shape}') squared_norm = (input_tensor ** 2).sum(-1, keepdim=True) print(f'squared_norm.shape: {squared_norm.shape}') ``` Print out : ``` input_tensor.shape: torch.Size([100, 1, 10, 16,...

NeVerDieKinG