Classification for Landslide Detection by ITU
1 000 CHF
Completed (~1 year ago)
Classification
Earth Observation
Python
GIS
Computer Vision
Machine Learning
Deep Learning
1028 joined
303 active
Start
17 Apr 25
Close
04 Aug 25
Reveal
04 Aug 25
Sharing Methods and Insights
8 Oct 2025, 15:39 · 1
Hello everyone,
first of all, big congratulations to the winners of this challenge! 🎉 I thought it would be nice to share some insights from this challenge (just as Koleshjr did) so we can all learn together as a community.
Here is my solution: https://github.com/PhilipJanz/Landslide-Image-Classification
It’s a fairly lightweight CNN-based model that takes RGB and SAR as separate input domains, merging them halfway through the network. I tried hard to use pre-trained models, but I couldn’t beat my own model trained from scratch. If anyone managed to find a well-suited pre-trained model, I’d love to hear about it! 🙂
@ geometry dash lite Really interesting approach, especially separating RGB and SAR before merging them later in the network. It’s always surprising how custom lightweight architectures can outperform pre-trained models in specialized datasets like this. Thanks for sharing the repository and the details behind your solution. I’m also curious whether anyone found a pre-trained backbone that handled SAR data particularly well during this challenge.