torch.stack 实现通道信息交叉混合

m0_61899108

已于 2022-11-06 00:43:18 修改

阅读量483

点赞数

分类专栏：代码阅读系列文章标签： python 深度学习开发语言

于 2022-11-05 23:59:47 首次发布

本文链接：https://blog.csdn.net/m0_61899108/article/details/127711098

版权

代码阅读系列专栏收录该内容

37 篇文章 4 订阅

订阅专栏

From

论文：TINYCD: A (Not So) Deep Learning Model For Change Detection

[2207.13159] TINYCD: A (Not So) Deep Learning Model For Change Detection (arxiv.org)

代码：

GitHub - AndreaCodegoni/Tiny_model_4_CD: Official implementation of TINYCD: A (NOT SO) DEEP LEARNING MODEL FOR CHANGE DETECTION

实现通道信息交叉混合，再分组卷积

代码实现

# https://github.com/AndreaCodegoni/Tiny_model_4_CD/blob/main/models/layers.py

class MixingBlock(nn.Module):
    def __init__(
        self,
        ch_in: int,
        ch_out: int,
    ):
        super().__init__()
        self._convmix = nn.Sequential(
            nn.Conv2d(ch_in, ch_out, 3, groups=ch_out, padding=1),
            nn.PReLU(),
            nn.InstanceNorm2d(ch_out),
        )

    def forward(self, x, y):
        # Packing the tensors and interleaving the channels:
        mixed = torch.stack((x, y), dim=2)        
        mixed = torch.reshape(mixed, (x.shape[0], -1, x.shape[2], x.shape[3]))

        # Mixing:
        return self._convmix(mixed)

if __name__ == '__main__':
    inputx = torch.randn(1,2,5,5)
    inputy = torch.randn(1,2,5,5)

    model = MixingBlock(2,2)
    output = model(inputx,inputy)

    print(output)