Reducing reliance on very high-resolution images and large models
Develop methods that allow APT to deliver meaningful acceleration and accuracy preservation at lower resolutions and on smaller model scales, thereby broadening its applicability beyond extremely high-resolution inputs and large architectures.
References
It also requires extremely high-resolution images and large models, making it an ideal application for our work.
— Accelerating Vision Transformers with Adaptive Patch Sizes
(Choudhury et al., 20 Oct 2025) in Conclusion, Limitations