Support for image generation with APT
Extend Adaptive Patch Transformers to image generation tasks by constructing mechanisms that enable variable patch sizing in generative models without degrading sample quality or efficiency.
References
Furthermore, while APT works for image understanding tasks, it does not currently support image generation.
— Accelerating Vision Transformers with Adaptive Patch Sizes
(Choudhury et al., 20 Oct 2025) in Conclusion, Limitations