GFNet

GFNet copied to clipboard

您好，感谢您的工作。我有两个问题想咨询。 1.我在有关序列任务上尝试了你的模型，我的序列长度为256，tensor为（B，256，512）输入到你的模型，我这里忽略了输入的意义，直接将序列256看作是2维信号 (即 H W 为16) ，强行让模型学习，但是奇怪的是结果却很好。我想问一下如何直接使用1D FFT对序列进行操作，代码如何编写？ 2.在进行2D FFT，输入的信号必须像文中的图片那样 H W是一样的吗？如何可以不一样，代码需要做那些修改？ 打扰了，期待您的回复！再次感谢！

Aerikervalid

Do you plan to update the code? (ft. TPAMI version)

2

Thanks to share your nice work. I have followed up on your art since the last year and I found out TPAMI paper which is enhanced from the NeurIPS version....

DoranLyong

The dimenson for Global Filter

1

import torch import torch.nn as nn import torch.fft class GlobalFilter(nn.Module): def __init__(self, dim, h=14, w=8): super().__init__() self.complex_weight = nn.Parameter(torch.randn(h, w, dim, 2, dtype=torch.float32) * 0.02) def forward(self, x): B, H,...

wangxinchao-bit

[Q] : can The Global filter Network use for Segmentation images

3

I read the paper, it was providing straight ideas about how does GFNT tackled the limitation of using Self-Attention with a Vision-Transformer in computations cost and complexity also, i would...

deep-matter

why I can only replicate the whole model on different gpu, not the model parallel that distributed part of the model to the gpu? I follow the instruction on the...

99sun99

Hello, i am new to pytorch and wanted to know how exactly the trained model could be used for prediction with images? thank you

kmg100

Ask about the reasoning speed of an fft filter: After using GlobalFilter, I found that this will greatly affect the speed of model reasoning, which is almost half of that...

ZZM11

About visualization

1

Thank you very much for your work, but when I realized the visualization of the frequency domain filter, the image is different from the one you provided. And I would...

iceCream-sh