keras-contrib
A deprecated repository for community-contributed Keras extensions like layers, activations, and loss functions.
What is keras-contrib?
Keras-contrib was an official repository for community-contributed extensions to the Keras deep learning framework. It provided additional layers, activations, loss functions, and optimizers that were not yet part of core Keras, allowing users to experiment with cutting-edge functionality while maintaining compatibility with standard Keras workflows.
Deep learning practitioners and researchers using Keras who need experimental or specialized model components beyond the core library's offerings.
It offered a curated collection of community-vetted extensions that seamlessly integrated with Keras, serving as both a testing ground for new ideas and a repository for useful but non-core functionality.
Overview
Keras community contributions
Use Cases
Best For
- Experimenting with advanced activation functions like PELU
- Adding specialized normalization layers to Keras models
- Testing community-developed loss functions for research
- Extending Keras with experimental optimizer variants
- Accessing archived Keras components not in the core library
- Learning about potential future Keras features through community contributions
Not Ideal For
- New projects requiring long-term maintenance and updates
- Teams needing seamless model loading without custom import steps
- Production environments demanding stable, supported extensions
- Users tightly integrated with modern TensorFlow 2.x and Keras ecosystems
Pros & Cons
Pros
Modules like PELU and GroupNormalization work directly with core Keras models, as shown in the example where they're added like standard layers without breaking workflows.
Served as a proving ground for tested contributions, with functionality validated for potential inclusion in core Keras, ensuring quality beyond random community code.
Provided early access to advanced components like Parametric Exponential Linear activations and specialized loss functions, filling gaps in core Keras at the time.
Acted as a repository for useful contributions that didn't fit Keras's minimalist paradigm, preserving community innovations for reference and niche use.
Cons
The README explicitly states it's deprecated with migration to TensorFlow Addons, meaning no bug fixes, updates, or security patches, making it risky for current projects.
Loading saved models requires importing each custom module and specifying custom_objects, adding overhead and potential errors compared to pure Keras models, as highlighted in the 'Gotcha' section.
Documentation is outdated and redirects to TensorFlow Addons, lacking guidance for current TensorFlow versions and leaving users to figure out compatibility issues on their own.
As an archived project, it lacks integration with newer Keras/TensorFlow features and community tools, potentially causing breakage in evolving deep learning workflows.
Frequently Asked Questions
Related Projects
keras-rlDeep Reinforcement Learning for Keras.
SpektralGraph Neural Networks with Keras and Tensorflow 2.
HyperasKeras + Hyperopt: A very simple wrapper for convenient hyperparameter optimization
TalosHyperparameter Experiments with TensorFlow and Keras
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.