wgcban / SemiCD

Revisiting Consistency Regularization for Semi-supervised Change Detection in Remote Sensing Images
https://www.wgcban.com/research#h.ar24vwqlm021
125 stars 19 forks source link

Problems with custom dataset #10

Open adesfoss opened 1 year ago

adesfoss commented 1 year ago

Hello, I'm working with your github repo and a custom dataset in semi-supervised mode and I'm having issues making it work. The program seems to have no issue whatsoever with the LEVIR dataset but when using my custom dataset, for some reason the IoU(change-ul) remains 0.00 (Loss and labeled change seem fine). I've tried playing around with the various decoder parameters to see if I could spot any variation in the metric with no encouraging results. Would you have any idea as to why that could be the case ? Thank you