prediction

[najsdh]

Hello, I'd like to ask about the code mentioned in your paper for using U-Net and Mask R-CNN for prediction. However, I only found HRNet in your code. Could you please guide me on where to find the code for using U-Net and Mask R-CNN for prediction? Thank you.

[HeavenswordG]

Hi @najsdh, in our research published in the ISPRS Journal, we initially utilized a hybrid approach combining U-Net and Mask R-CNN for building detection. However, upon transitioning our project to open-source, we discovered that directly implementing HRNet yielded more robust results across various datasets. Consequently, we have chosen to release only the HRNet code as our primary method for building segmentation. This approach has demonstrated a marginally superior performance compared to the U-Net and Mask R-CNN combination.

Thank you for being interested in our project.

[najsdh]

Then I would like to ask if you are willing to share the code of U-Net and Mask R-CNN before you, I would like to learn it, thank you very much

[HeavenswordG]

Then I would like to ask if you are willing to share the code of U-Net and Mask R-CNN before you, I would like to learn it, thank you very much

My code for U-Net and Mask R-CNN are in different Python environments and their versions are quite old. It is not easy to build up the same environment as mine, and the code is not organized. So I suggest that you can download U-Net code via MMSegmentation (https://github.com/open-mmlab/mmsegmentation), and Mask R-CNN code from AIcrowd (https://github.com/crowdAI/mapping-challenge-starter-kit and https://www.aicrowd.com/challenges/mapping-challenge), that's the code I used for Mask R-CNN training and prediction.