lironui/abcnet

The semantic segmentation of remote sensing images

ABCNet: Attentive Bilateral Contextual Network for Efficient Semantic Segmentation of Fine-Resolution Remote Sensing Images

⭐ Welcome to my HomePage ⭐

In this repository, we implement the Attentive Bilateral Contextual Network which contains a spatial path and a contextual path to fully capture the long-range relationships and fine-grained details in fine-resolution remote sensing images.

The detailed results can be seen in the ABCNet: Attentive Bilateral Contextual Network for Efficient Semantic Segmentation of Fine-Resolution Remote Sensing Images.

The training and testing code can refer to GeoSeg.

The related repositories include:

• MACU-Net->A modified version of U-Net.
• BANet->A Transformer-based segmentation network.
• MAResU-Net->A ResNet-based network with attention mechanism.
• Multi-Attention-Network->A network with multi kernel attention mechanism.

If our code is helpful to you, please cite:

R. Li, S. Zheng, C. Zhang, C. Duan, L. Wang, and P. M. Atkinson, "ABCNet: Attentive bilateral contextual network for efficient semantic segmentation of Fine-Resolution remotely sensed imagery," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 181, pp. 84-98, 2021/11/01/ 2021, doi: https://doi.org/10.1016/j.isprsjprs.2021.09.005.

Network:

Fig. 1. The overall architecture of ABCNet.

Result:

The result on the UAVid dataset can seen from here, where the user name is lironui or here and can be downloaded by this link:

Method	building	tree	clutter	road	vegetation	static car	moving car	human	mIoU
MSD	79.8	74.5	57.0	74.0	55.9	32.1	62.9	19.7	57.0
Fast-SCNN	75.7	71.5	44.2	61.6	43.4	19.5	51.6	0.0	45.9
BiSeNet	85.7	78.3	64.7	61.1	77.3	63.4	48.6	17.5	61.5
SwiftNet	85.3	78.2	64.1	61.5	76.4	62.1	51.1	15.7	61.1
ShelfNet	76.9	73.2	44.1	61.4	43.4	21.0	52.6	3.6	47.0
ABCNet	86.4	79.9	67.4	81.2	63.1	48.4	69.8	13.9	63.8

Fig. 2. The experimental results on the UAVid test set. The first column illustrates the input RGB images, the second column depicts the outputs of MSD and the third column shows the predictions of our ABCNet.

abcnet's Issues

求

大佬，全部代码合适开源吗？

代码训练有点小问题

您好，您提供的链接没有关于ABCNet的训练代码，所以我自己写了一个，但是感觉这个模型定义是不是有点问题

class ABCNet(nn.Module):
    def __init__(self, band, n_classes):
        super(ABCNet, self).__init__()
        self.name = 'ABCNet'
        self.cp = ContextPath()
        self.sp = SpatialPath()
        self.fam = FeatureAggregationModule(256, 256)
        self.conv_out = Output(256, 256, n_classes, up_factor=8)
        if self.training:
            self.conv_out16 = Output(128, 64, n_classes, up_factor=8)
            self.conv_out32 = Output(128, 64, n_classes, up_factor=16)
        self.init_weight()

    def forward(self, x):
        H, W = x.size()[2:]
        feat_cp8, feat_cp16, feat_cp32 = self.cp(x)
        feat_sp = self.sp(x)
        feat_fuse = self.fam(feat_sp, feat_cp8)

        feat_out = self.conv_out(feat_fuse)
        if self.training:
            feat_out16 = self.conv_out16(feat_cp16)
            feat_out32 = self.conv_out32(feat_cp32)
            return feat_out, feat_out16, feat_out32
        # feat_out = feat_out.argmax(dim=1)
        return feat_out

这个网络训练的时候，进self.training的判断的时候，feat_cp16的大小是[4, 256, 32, 32]，有256个通道。但是self.conv_out16定义的时候输出通道是128。feat_cp32的大小是[4, 512, 16, 16]，有512个通道。但是self.conv_out16定义的时候输出通道还是128.（我网络进去的图的大小是[4, 3, 512, 512]，是我哪操作不对吗？

请问怎么对uavid测试进行mlou评估的

大神，可以分享更加全面一些的代码学习吗

您好，我是一名语义分割的萌新，最近在网上阅读到大神的论文，下载调试时，总是会出现各种问题，想请问下，您有完整的代码可以分享吗，包括数据输入的模块

您好，请问方便分享一下config文件和损失函数的定义吗

自己试着改了config和损失函数，miou只有86.2，与论文中的88.56差距有些大，在您分享的GeoSeg中也没有找到相应的文件。请问方便分享吗，谢谢！

ABCNet损失函数

可以分享一下您abcnet的损失函数吗