This repository is a curated collection of research papers and code related to Masked Autoencoders (MAE) and similar masked modeling techniques, primarily focusing on their application in computer vision, but also extending to audio, graphs, and other domains. It serves as a valuable resource for researchers and practitioners interested in self-supervised learning, representation learning, and the advancements in Transformer-based architectures for various data modalities.

How It Works

The core concept revolves around masking a significant portion of input data (e.g., image patches, audio segments, graph nodes) and training a model, typically a Transformer, to reconstruct the missing parts. This self-supervised approach allows models to learn rich, generalizable representations from unlabeled data, significantly advancing the state-of-the-art in various downstream tasks. The advantage lies in its simplicity, scalability, and effectiveness in learning robust features without requiring extensive labeled datasets.

Quick Start & Requirements

This repository is a literature collection and does not have a direct installation or execution command. Users are directed to individual project pages or GitHub repositories linked for each paper to find specific setup instructions, dependencies (e.g., Python, PyTorch, TensorFlow, CUDA), and usage examples.

Highlighted Details

• Comprehensive coverage of MAE and related masked modeling techniques across Vision, Audio, and Graph domains.
• Links to numerous research papers with direct access to code repositories for practical implementation.
• Highlights advancements in self-supervised learning and representation learning for diverse data types.
• Includes papers exploring theoretical understandings and novel applications of masked modeling.

Maintenance & Community

This is a community-driven collection, with contributions from various researchers. The primary maintainer is EdisonLeeeee. Further community engagement and updates would depend on the activity of the contributors and the broader research community.

Licensing & Compatibility

The repository itself is a collection of links and does not have a specific license. Each linked paper and its associated code repository will have its own licensing terms, which users must adhere to. Compatibility for commercial use or closed-source linking will vary by project.

Limitations & Caveats

As a curated list, this repository does not provide a unified codebase or framework. Users must navigate to individual project repositories for each paper to obtain code, set up environments, and understand specific implementation details and limitations. The rapid pace of research means the list may require continuous updates to remain exhaustive.