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?
The given information is not enough to understand the issue because the train and test data will be different to train the model and evaluate. Please provide us some more information on the issue or provide us the code details to replicate the error and find the issue. Thank you.
