Investigating Focal and Dice Loss for the Kaggle 2018 Data Science Bowl

This post details my experiments and implementations with three important loss functions for the Kaggle 2018 data science bowl, and compares their effects on a simplified implementation of U-Net.

The loss functions I investigate are binary cross entropy (referred to as “nll” in the notebook because my initial version used the related NLLLoss instead of BCE), the soft-dice loss (introduced in “V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation” and generally considered to be useful for segmentation problems), and the focal loss.

View full post here

In the event you would want to cite this blog post, you could use this template:

@electronic{ecoffet2018investigating_focal_and_dice_loss_for_the_kaggle_2018_data_science_bowl,
  title={Investigating Focal and Dice Loss for the Kaggle 2018 Data Science Bowl},
  author={Ecoffet, Adrien},
  url = {https://medium.com/becoming-human/investigating-focal-and-dice-loss-for-the-kaggle-2018-data-science-bowl-65fb9af4f36c},
  year={2018}
}