-2

I have built a U-net model for image segmentation of 3-channel remote sensing images. I have two classes that look very similar and are hard to distinguish in the images without extra context. In this case, extra context would be the distance to something (i.e., distinguishing very similar looking buildings based on their proximity to a main road).

To add more context I've created an extra channel of values representing the distance to the nearest road, to try to improve segmentation results, but haven't got the results I expected.

Aside from my assumption of extra context being incorrect, is there something I'm missing? Are CNNs able to understand spatial relationships in images?

Christoph Rackwitz
  • 11,317
  • 4
  • 27
  • 36
Ellio
  • 117
  • 8
  • the distance channel seems to be a good idea. you should keep that. CNNs can understand spatial relationships as long as they're within the receptive field. that depends on the number of convolutions and pooling layers. the network you have can probably not "see" far enough. -- you might wanna check out https://ai.stackexchange.com/ or http://stats.stackexchange.com/ – Christoph Rackwitz Sep 01 '23 at 18:33

0 Answers0