# Not All Images are Worth 16x16 Words: Dynamic Transformers for Efficient Image Recognition

arXiv: https://arxiv.org/abs/2105.15075 · HTML: https://mlsanity.com/paper/2105.15075/

## Key facts

- Cited by 144 papers in the map (data through 14 Oct 2026).
- Builds on 51 of its 62 references that are in the map.
- First posted on arXiv 31 May 2021.

## Builds on (most cited first, 30 of 51)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://mlsanity.com/paper/2103.14030/) (2021)
- [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://mlsanity.com/paper/1905.11946/) (2019)
- [Layer Normalization](https://mlsanity.com/paper/1607.06450/) (2016)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [mixup: Beyond Empirical Risk Minimization](https://mlsanity.com/paper/1710.09412/) (2017)
- [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://mlsanity.com/paper/1804.07461/) (2018)
- [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://mlsanity.com/paper/1801.04381/) (2018)
- [Training data-efficient image transformers & distillation through attention](https://mlsanity.com/paper/2012.12877/) (2020)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features](https://mlsanity.com/paper/1905.04899/) (2019)
- [RandAugment: Practical automated data augmentation with a reduced search space](https://mlsanity.com/paper/1909.13719/) (2019)
- [Searching for MobileNetV3](https://mlsanity.com/paper/1905.02244/) (2019)
- [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://mlsanity.com/paper/2102.12122/) (2021)
- [Random Erasing Data Augmentation](https://mlsanity.com/paper/1708.04896/) (2017)
- [ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices](https://mlsanity.com/paper/1707.01083/) (2017)
- [ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design](https://mlsanity.com/paper/1807.11164/) (2018)
- [Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet](https://mlsanity.com/paper/2101.11986/) (2021)
- [Deep Networks with Stochastic Depth](https://mlsanity.com/paper/1603.09382/) (2016)
- [CvT: Introducing Convolutions to Vision Transformers](https://mlsanity.com/paper/2103.15808/) (2021)
- [Multiscale Vision Transformers](https://mlsanity.com/paper/2104.11227/) (2021)
- [CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification](https://mlsanity.com/paper/2103.14899/) (2021)
- [Going deeper with Image Transformers](https://mlsanity.com/paper/2103.17239/) (2021)
- [Transformer in Transformer](https://mlsanity.com/paper/2103.00112/) (2021)
- [BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks](https://mlsanity.com/paper/1709.01686/) (2017)
- [ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases](https://mlsanity.com/paper/2103.10697/) (2021)

## Built on by (most cited first, 30 of 144)

- [Least-to-Most Prompting Enables Complex Reasoning in Large Language Models](https://mlsanity.com/paper/2205.10625/) (2022)
- [Measuring and Narrowing the Compositionality Gap in Language Models](https://mlsanity.com/paper/2210.03350/) (2022)
- [Vision Transformer with Deformable Attention](https://mlsanity.com/paper/2201.00520/) (2022)
- [FLatten Transformer: Vision Transformer using Focused Linear Attention](https://mlsanity.com/paper/2308.00442/) (2023)
- [GSVA: Generalized Segmentation via Multimodal Large Language Models](https://mlsanity.com/paper/2312.10103/) (2023)
- [FlexiViT: One Model for All Patch Sizes](https://mlsanity.com/paper/2212.08013/) (2022)
- [A Survey of Visual Transformers](https://mlsanity.com/paper/2111.06091/) (2021)
- [DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution](https://mlsanity.com/paper/2411.02359/) (2024)
- [CF-ViT: A General Coarse-to-Fine Method for Vision Transformer](https://mlsanity.com/paper/2203.03821/) (2022)
- [Dynamic Token Pruning in Plain Vision Transformers for Semantic Segmentation](https://mlsanity.com/paper/2308.01045/) (2023)
- [Dynamic Diffusion Transformer](https://mlsanity.com/paper/2410.03456/) (2024)
- [Which Tokens to Use? Investigating Token Reduction in Vision Transformers](https://mlsanity.com/paper/2308.04657/) (2023)
- [Adaptive Rotated Convolution for Rotated Object Detection](https://mlsanity.com/paper/2303.07820/) (2023)
- [Rank-DETR for High Quality Object Detection](https://mlsanity.com/paper/2310.08854/) (2023)
- [A Survey on Transformer Compression](https://mlsanity.com/paper/2402.05964/) (2024)
- [EfficientTrain: Exploring Generalized Curriculum Learning for Training Visual Backbones](https://mlsanity.com/paper/2211.09703/) (2022)
- [AIM: Adaptive Inference of Multi-Modal LLMs via Token Merging and Pruning](https://mlsanity.com/paper/2412.03248/) (2024)
- [Glance and Focus Networks for Dynamic Visual Recognition](https://mlsanity.com/paper/2201.03014/) (2022)
- [AdaFocus V2: End-to-End Training of Spatial Dynamic Networks for Video Recognition](https://mlsanity.com/paper/2112.14238/) (2021)
- [ResFormer: Scaling ViTs with Multi-Resolution Training](https://mlsanity.com/paper/2212.00776/) (2022)
- [Expediting Large-Scale Vision Transformer for Dense Prediction without Fine-tuning](https://mlsanity.com/paper/2210.01035/) (2022)
- [FGAHOI: Fine-Grained Anchors for Human-Object Interaction Detection](https://mlsanity.com/paper/2301.04019/) (2023)
- [Fast yet Safe: Early-Exiting with Risk Control](https://mlsanity.com/paper/2405.20915/) (2024)
- [Making Vision Transformers Efficient from A Token Sparsification View](https://mlsanity.com/paper/2303.08685/) (2023)
- [Improving Transformers with Probabilistic Attention Keys](https://mlsanity.com/paper/2110.08678/) (2021)
- [PatchDropout: Economizing Vision Transformers Using Patch Dropout](https://mlsanity.com/paper/2208.07220/) (2022)
- [ALGM: Adaptive Local-then-Global Token Merging for Efficient Semantic Segmentation with Plain Vision Transformers](https://mlsanity.com/paper/2406.09936/) (2024)
- [Scaling Vision Pre-Training to 4K Resolution](https://mlsanity.com/paper/2503.19903/) (2025)
- [Stitchable Neural Networks](https://mlsanity.com/paper/2302.06586/) (2023)
- [SGAT4PASS: Spherical Geometry-Aware Transformer for PAnoramic Semantic Segmentation](https://mlsanity.com/paper/2306.03403/) (2023)

## Abstract (arXiv)

Vision Transformers (ViT) have achieved remarkable success in large-scale image recognition. They split every 2D image into a fixed number of patches, each of which is treated as a token. Generally, representing an image with more tokens would lead to higher prediction accuracy, while it also results in drastically increased computational cost. To achieve a decent trade-off between accuracy and speed, the number of tokens is empirically set to 16x16 or 14x14. In this paper, we argue that every image has its own characteristics, and ideally the token number should be conditioned on each individual input. In fact, we have observed that there exist a considerable number of "easy" images which can be accurately predicted with a mere number of 4x4 tokens, while only a small fraction of "hard" ones need a finer representation. Inspired by this phenomenon, we propose a Dynamic Transformer to automatically configure a proper number of tokens for each input image. This is achieved by cascading multiple Transformers with increasing numbers of tokens, which are sequentially activated in an adaptive fashion at test time, i.e., the inference is terminated once a sufficiently confident prediction is produced. We further design efficient feature reuse and relationship reuse mechanisms across different components of the Dynamic Transformer to reduce redundant computations. Extensive empirical results on ImageNet, CIFAR-10, and CIFAR-100 demonstrate that our method significantly outperforms the competitive baselines in terms of both theoretical computational efficiency and practical inference speed. Code and pre-trained models (based on PyTorch and MindSpore) are available at https://github.com/blackfeather-wang/Dynamic-Vision-Transformer and https://github.com/blackfeather-wang/Dynamic-Vision-Transformer-MindSpore.