Petros Giannakopoulos
Petros Giannakopoulos
I gave it a shot, however I am not sure how the discounted reward R is supposed to be used and I also need to check if future and past...
I have tried implementing optimality tightening (see earlier post) but the results I get are also much worse than the paper's.
Hello, You are correct that the architecture can be used for any segmentation task. In fact, the U-Net architecture was first used for [biomedical image segmentation](https://arxiv.org/abs/1505.04597).
The limiting factor is that the model must be expressive enough to solve a task with an accuracy you deem acceptable. If the architecture doesn't perform as well as you...
Can't reproduce this issue. Is your Keras version up-to-date? If you remove validation from fit_generator do you still get the error?
Have you tried on another Python environment?
On my GTX 1070 each epoch takes ~150s for 128, ~250s for 256 , ~550s for 512 and ~1550s for 1024.
I got my 0.996 result with batch size = 2. Training becomes more unstable with small batch sizes and I think Batchnorm can't work at all with batch size =...
You can experiment with different kinds and amount of augmentations to find the best spot between convergence and avoiding overfitting. You can use the `u` parameter of an augmentation function...