focal-loss 0.0.4

TensorFlow implementation of focal loss.

TensorFlow implementation of focal loss [1]: a loss function generalizing binary cross-entropy loss that penalizes hard-to-classify examples.

The focal_loss package provides a function binary_focal_loss and a class BinaryFocalLoss that can be used as stand-in replacements for tf.keras.losses functions and classes, respectively.

# Typical tf.keras API usage
import tensorflow as tf
from focal_loss import BinaryFocalLoss

model = tf.keras.Model(...)
model.compile(
    optimizer=...,
    loss=BinaryFocalLoss(gamma=2),  # Used here like a tf.keras loss
    metrics=...,
)
history = model.fit(...)

Documentation is available at Read the Docs.

Installation

1. Make sure that a CPU or GPU version of TensorFlow 2.0 or later is installed (see this link for installation instructions).

2. The focal_loss package can be installed using the pip utility. For the latest version, install directly from the package’s GitHub page:

   pip install git+https://github.com/artemmavrin/focal-loss.git

   Alternatively, install a recent release from the Python Package Index (PyPI):

   pip install focal-loss

   Note. To install the project for development (e.g., to make changes to the source code), clone the project repository from GitHub and run make dev:

   git clone https://github.com/artemmavrin/focal-loss.git
   cd focal-loss
   # Optional but recommended: create a new Python virtual environment first
   make dev

   This will additionally install the requirements needed to run tests, check code coverage, and produce documentation.

References

[1]

T. Lin, P. Goyal, R. Girshick, K. He and P. Dollár. Focal loss for dense object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018. (DOI) (arXiv preprint)