Cookies help us deliver our services. By using our services, you agree to our use of cookies. Learn more

crowdAI is shutting down - please read our blog post for more information

Mapping Challenge

Building Missing Maps with Machine Learning


Which approach? R-CNN or U-net?

Posted by aglotero over 3 years ago

Hi folks!

I’m curious about your approach: u-net or r-cnn?

Regards, André

Posted by chenyiyong  over 3 years ago |  Quote

I am using R-CNN for now. But I am no sure which one will perform better?

Posted by aglotero  over 3 years ago |  Quote

My first poor LoL UNet: https://www.crowdai.org/528147398460

Posted by neptune.ml  about 3 years ago |  Quote

We are using U-net and I am pretty sure that it is the way to go. It seems that due to the metric that is used missing small buildings has a huge impact on the score. For what I know detection networks (and I presume Mask-R-CNN too) have trouble with small objects. I think the way we deal with those small buildings (and parts of buidlings on the edges) will decide the top spots.