# Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolution

**Authors:** Mostafa Dehghani, Basil Mustafa, Josip Djolonga, Jonathan Heek, Matthias Minderer, Mathilde Caron, Andreas Steiner, Joan Puigcerver, Robert Geirhos, Ibrahim Alabdulmohsin, Avital Oliver, Piotr Padlewski, Alexey Gritsenko, Mario Lučić, Neil Houlsby  
**arXiv:** [2307.06304](https://arxiv.org/abs/2307.06304)  
**Date:** 2023-07-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.06304v1/

## Key points

- The authors present NaViT (Native Resolution ViT), a model that utilizes Vision Transformers (ViTs) for processing inputs of arbitrary resolutions and aspect ratios.
- NaViT represents a departure from the standard input and modeling pipeline used by most computer vision models, which are typically CNN-designed.
- NaViT improves training efficiency for large-scale supervised and contrastive image-text pretraining.
- It can be efficiently transferred to standard tasks such as image and video classification, object detection, and semantic segmentation.
- NaViT leads to improved results on robustness and fairness benchmarks.
- The flexibility in handling varying input sequence lengths and resolutions makes NaViT a promising direction for ViTs.

## Summary

The authors of the paper, Mostafa Dehghani, Basil Mustafa, Josip Djolonga, Jonathan Heek, Matthias Minderer, Mathilde Caron, Andreas Steiner, Joan Puigcerver, Robert Geirhos, Ibrahim Alabdulmohsin, Avital Oliver, Piotr Padlewski Alexey Gritsenko Mario Lučić and Neil Houlsby present NaViT (Native Resolution ViT), a model that takes advantage of the flexible sequence-based modeling offered by Vision Transformers (ViTs) to process inputs of arbitrary resolutions and aspect ratios. This represents a departure from the standard input and modeling pipeline used by most computer vision models which are typically CNN-designed. NaViT is shown to improve training efficiency for large-scale supervised and contrastive image-text pretraining. Furthermore it can be efficiently transferred to standard tasks such as image and video classification object detection and semantic segmentation leading to improved results on robustness and fairness benchmarks. The flexibility in handling varying input sequence lengths and resolutions make NaViT a promising direction for ViTs.