SKNet

SKNet copied to clipboard

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [8, 64, 32, 32]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later.

2

Run sknet.py, got RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [8, 64, 32, 32]], which is output 0 of ReluBackward0,...

Jin-Xi

sorry,i think it's a easy problem.

fangzhouAlonso

The Global Average Pooling

2

https://github.com/pppLang/SKNet/blob/ef76d8dd87a3c12eac69169758ee90fd1e189372/sknet.py#L45 https://github.com/pppLang/SKNet/blob/ef76d8dd87a3c12eac69169758ee90fd1e189372/sknet.py#L46 Does the code use mean instead of GAP? Is the performance obvious?

jianjiandandande

我认为这个操作和paper里不符合，应该是如下： batch_size, ch = 5, 3 feat_a = torch.randn(batch_size, ch) feat_a.unsqueeze_(-1) feat_b = torch.randn(batch_size, ch) feat_b.unsqueeze_(-1) feat_tmp = torch.cat([feat_a, feat_b], dim=-1) feat_softmax = torch.softmax(feat_tmp, dim=-1) weight_a = feat_softmax[:, :, 0].squeeze()...

dangxusheng

RuntimeError: size mismatch, m1: [12288 x 3], m2: [1024 x 196] at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:290 Thanks for your help！

Lin-plax

Why no 'dilation'?

1

https://github.com/pppLang/SKNet/blob/ef76d8dd87a3c12eac69169758ee90fd1e189372/sknet.py#L24 Shouldn’t it be like this? `nn.Conv2d(features, features, kernel_size=3+i*2, stride=stride, padding=1+i, groups=G, dilation=1+i),`

ghost

LiDaiY

您好，问一下关于sknet.py的问题

3

SKUnit这个类是用来干嘛的。。。

Xuecheng-anhui

RuntimeError: Given input size: (1024x4x4). Calculated output size: (1024x0x0). Output size is too small

1

The following error occurred when I used cifiar-10 data to train SK network. Is it because the image size is too small (32 * 32)? Traceback (most recent call last):...

caisong1995

Great work. If I am not wrong, official implementation conv1 use kernel size=7, did you intentionally change it to kernel size=3? Thanks!

doitslow