hi i’m training a model unet++ with resnet18 as encoder that segments satellite images into 6 classes .after 200 epochs i get 82% accuracy without over-fitting but accuracy doesn’t increase . is there any way to increase accuracy for example using other encoder
Try using a deeper backbone like ResNet 32 or something. Also does the loss stagnate after a few epochs, or is the loss still decreasing?
thanks for your reply it has been 30 epoch that loss and accuracy dont change
If loss and accuracy both do not change, I guess that you should use a deeper backbone network. Try using deeper ResNet models or Inception ResNet models, they might help.