Different architectures have been tested here from basic to high level. An implementation of soft attention could be found here. The architectures used are
- Baseline CNN
- CNN with Attention
- Inception V4
- Unet
- Attention Unet
Follow this link to download the whole dataset.
run train.py
and train_ismail.py
files. The paths should be adjusted as required. The hyperparameters can be set in hparams.py
file