I have run lesson2 ipynb in Colab several times, but I couldn't get a good segmentation result though the training stage is normal(val loss 0.010, val acc 0.95). In some cases, the masks don't even exist. I don't know whether this is because the model is not good or because the dataset is very small, maybe both. I have also changed another dataset which has 566 images, however I still can't get a good mask.
Now maybe I should try other implementation of U-Net, but I still hope there are someone who can explain to me.
I have run lesson2 ipynb in Colab several times, but I couldn't get a good segmentation result though the training stage is normal(val loss 0.010, val acc 0.95). In some cases, the masks don't even exist. I don't know whether this is because the model is not good or because the dataset is very small, maybe both. I have also changed another dataset which has 566 images, however I still can't get a good mask.
Now maybe I should try other implementation of U-Net, but I still hope there are someone who can explain to me.