Keras Deep Learning Models

Deep Learning Models is a classic open-source collection of Keras implementations of popular deep learning architectures, created by François Chollet, the author of Keras. The repository ships both the model code and pre-trained weights files for well-known image classification networks, so you can instantiate a famous architecture in a few lines and immediately use it for inference or fine-tuning. It has earned about 7,400 stars on GitHub and remains a frequently cited reference for how these networks are structured.

Overview

The collection includes VGG16, VGG19, ResNet50, Inception v3 and a CRNN model for music tagging. All architectures were written to work with both the TensorFlow and Theano backends, following the image dimension ordering in your Keras configuration, and pre-trained ImageNet weights can be loaded automatically when a model is created. The repository is now deprecated in favor of the built-in keras.applications module, and its README directs new users there, but the standalone code and weights remain available and widely referenced. Last updated on GitHub in 2020, it is best understood as a stable, historical snapshot of the models that defined the deep learning era rather than an actively developed library.

Key Features

• Ready-to-use Keras code for VGG16, VGG19, ResNet50 and Inception v3 image classification models
• Pre-trained ImageNet weights that load automatically on model instantiation
• A CRNN implementation for music tagging alongside the vision models
• Compatible with both TensorFlow and Theano backends and configurable image ordering
• Clear, readable reference implementations useful for learning how the architectures work
• Permissive MIT license for reuse in research and teaching

💳 Pricing

Deep Learning Models is completely free and open source under the MIT license. There is no paid tier; note that the author now recommends the keras.applications module built into modern Keras for new projects.

Who Is It For

This repository is for students, researchers and engineers who want clean reference implementations of landmark convolutional networks, whether to study their architecture, reproduce classic results, or lift a proven model definition into an older Keras codebase.