Implementation of Unet on Keras with BatchNormalization and Tensorboard visualization. The code is written keeping in mind that it will be run on Kaggle Carvana Image challenge (https://www.kaggle.com/c/carvana-image-masking-challenge).
After running Unet 4 times(epoch = 4) on only 12 images, we get the following output -
left:image, middle:ground truth, right:unet segmentation output
Below are the losses at every batch, epoch for training and validation sets.
References -
- Kaggle Discussions / kernels
- Other Unet implementations of Unet on Github like -
- Tensorboard implementation for logging at -