lironui/macu-net

The semantic segmentation of remote sensing images

MACU-Net for Semantic Segmentation of Fine-Resolution Remotely Sensed Images

⭐ Welcome to my HomePage ⭐

In this repository, we incorporate multi-scale features generated by different layers of U-Net and design a multi-scale skip connected and asymmetric-convolution-based U-Net (MACU-Net), for segmentation using fine-resolution remotely sensed images.

The detailed results can be seen in the MACU-Net for Semantic Segmentation of Fine-Resolution Remotely Sensed Images.

The training and testing code can refer to GeoSeg.

If our code is helpful to you, please cite:

R. Li, C. Duan, S. Zheng, C. Zhang and P. M. Atkinson, "MACU-Net for Semantic Segmentation of Fine-Resolution Remotely Sensed Images," in IEEE Geoscience and Remote Sensing Letters, doi: 10.1109/LGRS.2021.3052886.

Requirements：

numpy >= 1.16.5
PyTorch >= 1.3.1
sklearn >= 0.20.4
tqdm >= 4.46.1
imageio >= 2.8.0

Datasets:

• GID
• WHDLD

Note: We select 15 images contained in GID, which cover the whole six categories:

GF2_PMS1__L1A0000647767-MSS1
GF2_PMS1__L1A0001064454-MSS1
GF2_PMS1__L1A0001348919-MSS1
GF2_PMS1__L1A0001680851-MSS1
GF2_PMS1__L1A0001680853-MSS1
GF2_PMS1__L1A0001680857-MSS1
GF2_PMS1__L1A0001757429-MSS1
GF2_PMS2__L1A0000607681-MSS2
GF2_PMS2__L1A0000635115-MSS2
GF2_PMS2__L1A0000658637-MSS2
GF2_PMS2__L1A0001206072-MSS2
GF2_PMS2__L1A0001471436-MSS2
GF2_PMS2__L1A0001642620-MSS2
GF2_PMS2__L1A0001787089-MSS2
GF2_PMS2__L1A0001838560-MSS2

You can download datasets and prepare the files to ./datasets folder. Note:

Transfer the lable images form RGB Format to Grey-scale Map.
Augment the training set by the technology mentioned in our paper.

Network:

Fig. 1. Comparison of (a) U-Net, (b) U-Net++， and proposed (c) MACU-Net 3+. The depth of each node is presented below the circle.

Results:

Fig. 2. Visualization of results on the WHDLD dataset (the left) and the GID dataset (the right).

macu-net's Issues

Dataset

Hi, Thanks for your sharing.
You wrote, "Transfer the label images from RGB Format to Grey-scale Map". Although I follow this rule for the WHDLD dataset, the prediction results are not consistent. Could you explain this condition in more detail?

*dataset.py
data_series.append(imageio.imread(x_batch[i]) / 255.0) >> for scale [0,1]
label_series.append(imageio.imread(y_batch[i]) - 1) >> ???

请求

您好，
请问是否方便提供一下test脚本代码,谢谢

bug

Hello, your model reported an error during training. Can you provide a complete code?

代码和数据集

请问代码和数据集什么时候发布啊。可不可以先把数据集发布了

bug

Hello, your model reported an error during training. Can you provide a complete code?

Which model is implementation of MACU-Net?

Hi, there are so many codes in model directory which one is the MACU-Net?

Visual Results

I am working on a GID dataset for 5 and 15 classes but have an issue about the visualization of the output's color scheme as it is not giving the right colors of classes.
Can you please tell me how you applied the color scheme and in which order.

Thank you

可视化

请问有可视化得代码吗。我的结果可是有点糟糕