This repository serves as a curated list of recent research papers focused on autoregressive visual generation, a technique for creating images sequentially. It targets researchers and practitioners in computer vision and generative AI, providing a centralized resource for staying abreast of advancements in this rapidly evolving field.

How It Works

The repository tracks papers that employ autoregressive models, which generate images by predicting pixels or tokens one after another. This approach contrasts with diffusion models and GANs, offering potential advantages in controllability and interpretability by leveraging the sequential prediction capabilities of large language models (LLMs) adapted for visual data.

Quick Start & Requirements

This repository is a curated list of papers and does not have a direct installation or execution command. The papers themselves may have associated codebases with varying requirements.

Highlighted Details

• Focuses on autoregressive models, including those leveraging vector quantization (VQ-GAN variants, VAR) and masked image modeling (BEiT).
• Includes papers exploring the integration of autoregressive models with LLMs for multimodal generation.
• Features research on tokenization strategies, which are crucial for enabling autoregressive generation with discrete visual representations.
• Covers a wide range of recent publications from major conferences like CVPR, NeurIPS, and ICLR, as well as arXiv preprints.

Maintenance & Community

The repository is maintained by lxa9867. It also highlights the maintenance of the XQ-GAN framework. No specific community links (Discord, Slack) or roadmap are provided in the README.

Licensing & Compatibility

The repository itself is not a software project with a license. The linked papers and any associated codebases will have their own respective licenses.

Limitations & Caveats

This is a list of papers, not a runnable framework. Users must find and evaluate the codebases for individual papers to assess their functionality, performance, and specific requirements. The rapid pace of research means the list may not be exhaustive or immediately updated with the very latest publications.