Autoregressive-Models-in-Vision-Survey

Why we included this project

This is a peer-reviewed survey companion that organizes the recent wave of autoregressive models in vision, and it will save researchers and engineers a lot of time they would otherwise spend digging through papers. The repo is a taxonomy-driven index covering image, video, 3D, motion, point-cloud, and multimodal generation, with each entry linking to the original paper and code when it exists. Because it is built around a TMLR 2025 survey, the structure reflects an actual academic review rather than an ad-hoc tag cloud, and the detailed table of contents lets you jump straight to a sub-topic like text-to-image editing or embodied AI. Teams scoping a new generative project can use it to spot the dominant autoregressive approaches, benchmark papers, and acceleration techniques before committing to a method. Just keep in mind this is a reading and reference resource, not deployable software, so treat it as a starting point for literature review rather than something to run in production.