refrakt-core 0.1.0

A modular library for training, benchmarking, and experimenting with deep vision models.

refrakt-core (previously Re-Implementation)

Current re-implementations are:

Paper	Status
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale	Completed
Deep Residual Learning for Image Recognition	Completed
Autoencoders	Completed
Swin Transformer: Heirarchial Vision Transformer using Shifted Windows	Completed
Attention Is All You Need	Completed
A ConvNet for the 2020s	Completed
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network	Completed
A Simple Framework for Contrastive Learning of Visual Representations	Completed
Emerging Properties in Self-Supervised Vision Transformers	Formulation
Masked Autoencoders Are Scalable Vision Learners	Formulation
Masked Siamese Networks for Label-Efficient Learning	Formulation
Lagrangian Neural Networks	Formulation

Contributions:

These are my personal implementations in order to educate myself. That being said, if there are any issues with the code, such as incorrect math, not enough comments or documentation, or poor modularity, please create an issue so I can review and make changes. Pull requests must be the last resort.

Licensing:

This repository is under the MIT License. See the LICENSE file for more details.