Hi StackExchange community I am working on to train a Unet for satellite change segmentation. My dataset consists of images(before change),images(after change) and the corresponding change segmentation ground truth mask. Right now I am concatenating both the satellite images on channel axis and passing to Unet also I am changing the pixel values of mask which consists of [0,1,254,255] to only [0,1].I am using dice coefficient and dice loss to monitor the performance but my dice coefficient value is not increasing neither the model is able to segment properly .Any suggestions on what wrong I am doing and how can I improve the performance. Or any other approach that I should use in order to do satellite change segmentation.

Thank you very much


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.