Attention guides faster accurate image creation with less waste
ATLAS: Attention-Guided Test-Time Latent Selection for Compositional Visual Autoregressive Generation
Computer Vision and Pattern Recognition
Summary
When computers draw pictures from descriptions, they sometimes mix up details or positioning. The authors created ATLAS to help the computer focus on the parts it pays attention to while drawing, choosing better paths early and avoiding wasted effort. This method finds promising options quickly by checking partial results without finishing them all, saving time and memory. Testing shows ATLAS improves picture quality and uses less computing power when making images step-by-step.
What this means in practice
- •For multimedia system engineers: Improve image generation tools to produce more accurate composite pictures with reduced computation time by guiding search using internal attention signals.
- •For graphics software developers: Optimize rendering pipelines for generative art by selectively expanding promising image generation candidates before full decoding, saving GPU resources.
Authors
Hossein Shahabadi, Raya Namazi, Mahdieh Soleymani Baghshah
Abstract
Visual autoregressive models produce convincing images but can misassign attributes and spatial relations. Searching multiple generation paths can improve composition, yet completing every candidate is costly. We introduce ATLAS, a training-free framework that uses the frozen generator's internal attention to guide search before final images are available. Within-prompt diagnostics identify attention-derived scores that predict final compositional preferences on Infinity, with category-dependent transfer to Switti. ATLAS uses these on-the-fly signals to progressively concentrate computation on promising trajectories. It maintains multiple coarse alternatives, prunes candidates without decoding at intermediate scales, and reserves image-space evaluation for a final choice among partial trajectories. ATLAS scales effectively on both backbones, achieving higher mean scores on the evaluated T2I-CompBench and GenEval categories as more computation is devoted to search. At the largest tested budget, the targeted T2I-CompBench mean improves by over 10% relative to each backbone. Separate profiles show over 50% less generation time than completing and evaluating all candidates at equal initial counts, alongside lower peak allocated GPU memory. Relative time savings are larger at the higher tested candidate count. Together, these results support scalable compositional search: additional sampling expands exploration, while attention-guided selection limits continuation cost without parameter updates or gradient-based optimization.