I used the UNet model to do the satellite image classification, it works well and helps me a lot. But recently, I found that it is difficult to get a stable result. For example, I used the same input samples to train the model, but the results will be different.
The train and loss curve was attached. Did anyone have the same problems?
