RepLKNet-pytorch

RepLKNet-pytorch copied to clipboard

When applying model.structural_reparam(), there is a need to add padding so that the output size can match, see #48. Besides, I find that the kernel_size, stride, padding, and dilation may...

liyiersan

No module named 'timm.optim.novograd‘

3

File "D:\A_File\Project\RepLKNet-pytorch-main\optim_factory.py", line 19, in from timm.optim.novograd import NovoGrad ModuleNotFoundError: No module named 'timm.optim.novograd'

guyunsu128

DepthWiseConv2dImplicitGEMM has no 'padding' class attribute(actually zero)

9

When I ran directly `replknet.py` with using `DepthWiseConv2dImplicitGEMM`, I got this error: ``` RuntimeError: The size of tensor a (56) must match the size of tensor b (26) at non-singleton...

zzzc18

Instructions about how to install 19_large_depthwise_conv2d_torch_extension

1

Thanks for the author's nice work. I met some problems when installing large_depthwise_conv2d_torch_extension. However, I finally succeeded in the following ways. Environment: both pytorch1.12 + python3.9+cuda11.3 and pytorch1.10+python3.8+cuda11.3 on Ubuntu...

liyiersan

ModuleNotFoundError: No module named '_depthwise_conv2d_implicit_gemm_C'

3

Hi, thanks for your nice work. I encountered a problem when I ran `python depthwise_conv2d_implicit_gemm.py`. I got `ModuleNotFoundError: No module named '_depthwise_conv2d_implicit_gemm_C'`. Could you please tell me how I can...

XiaxiaBubble

ERF is smaller after changing the input image size

6

i changed the input size to 224×224，and compare RepLKNet with resnet101, it seems that resnet101 is larger than RepLKNet-13, left is RepLKNet-13, right is resnet101 

sleepat-11

Using a transformer type model to calculate ERF yields very poor results

1

864530979

I try to increase the -1 dim ,the big kernal becomes more and more slower than conv2d

1

if torch.cuda.is_available(): x = torch.randn(64, 384, 256, 31).cuda() m1 = DepthWiseConv2dImplicitGEMM(384, 31, bias=False).cuda() m2 = nn.Conv2d(384, 384, 31, padding=31 // 2, bias=False, groups=384).cuda() m2.load_state_dict(m1.state_dict()) with torch.cuda.amp.autocast(True): import time t1 =...

Marsjunwang

Hi, I tried to use structure re-param method in 3D segmentation task. The same prediction can be achieved using the 333+111+identity => 333. But it fails at 777+333+identity =>777. How...

TravisL24

# Welcome update to OpenMMLab 2.0 I am Vansin, the technical operator of OpenMMLab. In September of last year, we announced the release of OpenMMLab 2.0 at the World Artificial...

vansin