LeViT: a Vision Transformer in ConvNet’s Clothing for Faster Inference
Abstract
We design a family of image classification architectures that optimize the trade-off between accuracy and efficiency in a high-speed regime. Our work exploits recent findings in attention-based architectures, which are competitive on highly parallel processing hardware. We revisit principles from the extensive literature on convolutional neural networks to apply them to transformers, in particular activation maps with decreasing resolutions. We also introduce the attention bias, a new way to integrate positional information in vision transformers.As a result, we propose LeViT: a hybrid neural network for fast inference image classification. We consider different measures of efficiency on different hardware platforms, so as to best reflect a wide range of application scenarios. Our extensive experiments empirically validate our technical choices and show they are suitable to most architectures. Overall, LeViT significantly outperforms existing convnets and vision transformers with respect to the speed/accuracy tradeoff. For example, at 80% ImageNet top-1 accuracy, LeViT is 5 times faster than EfficientNet on CPU. We release the code at https://github.com/facebookresearch/LeViT.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
MSAs closer to the last layer significantly improve predictive performance.
“MSAs closer to the last layer significantly improve predictive performance (Graham et al., 2021; Dai et al., 2021).”
From How Do Vision Transformers Work? (2022) · cites this paper as Graham et al., 2021
According to studies on ViT, the later layers use high-level information, and the early layers exploit low-level information.
“According to studies on ViT(Graham et al., 2021;Dai et al., 2021; Park & Kim, 2022b), the later layers use high-level information, and the early layers exploit low-level information.”
From What Do Self-Supervised Vision Transformers Learn? (2023) · cites this paper as Graham et al., 2021
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · see citing works
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 48,293 · see citing works
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · see citing works
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818 · see citing works
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Tan, M., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1905.11946 · Cited by 17,072 · see citing works
Backpropagation Applied to Handwritten Zip Code Recognition
LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989) · Neural Computation · doi:10.1162/neco.1989.1.4.541 · Cited by 11,891 · see citing works
Wang, X., Girshick, R., Gupta, A., & He, K. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00813 · Cited by 11,596 · see citing works
Howard, A., Sandler, M., Chen, B., Wang, W., Chen, L.-C., Tan, M., Chu, G., Vasudevan, V., Zhu, Y., Pang, R., et al. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00140 · Cited by 9,708 · see citing works
Zagoruyko, S., & Komodakis, N. (2016) · Procedings of the British Machine Vision Conference 2016 · doi:10.5244/c.30.87 · Cited by 7,244 · see citing works
Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions
Wang, W., Xie, E., Li, X., Fan, D.-P., Song, K., Liang, D., Lu, T., Luo, P., & Shao, L. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00061 · Cited by 5,259 · cited 4× · see citing works
“Similar to PVT, its design relies on re-tokenization of the output after each layer by aggregating the neighboring tokens such number of tokens are progressively reduced.”
BackgroundMethodInfluentialInception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. (2017) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v31i1.11231 · Cited by 4,636 · see citing works
Residual Attention Network for Image Classification
Wang, F., Jiang, M., Qian, C., Yang, S., Li, C., Zhang, H., Wang, X., & Tang, X. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.683 · Cited by 4,073 · see citing works
Convolutional Sequence to Sequence Learning
Gehring, J., Auli, M., Grangier, D., Yarats, D., & Dauphin, Y. N. (2017) · arXiv · doi:10.48550/arxiv.1705.03122 · Cited by 2,850 · cited 1× · see citing works
“Other works have replaced them with parametric encoding [33] or adopt Fourier-based kernelized versions [22].”
MethodTraining Vision Transformers for Image Retrieval
El-Nouby, A., Neverova, N., Laptev, I., & Jégou, H. (2021) · arXiv · doi:10.48550/arxiv.2102.05644 · Cited by 127 · see citing works
Self-citation
Flagged first, then by citations
Cited by
Sorted by citations
Pvt v2: Improved baselines with pyramid vision transformer
Wang, W., Xie, E., Li, X., Fan, D.-P., Song, K., Liang, D., Lu, T., Luo, P., & Shao, L. (2022) · Computational Visual Media · doi:10.1007/s41095-022-0274-8 · Cited by 2,439 · see citing works
EfficientViT: Memory Efficient Vision Transformer with Cascaded Group Attention
Liu, X., Peng, H., Zheng, N., Yang, Y., Hu, H., & Yuan, Y. (2023) · 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52729.2023.01386 · Cited by 956 · see citing works
SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
Xie, E., Wang, W., Yu, Z., Anandkumar, A., Alvarez, J. M., & Luo, P. (2021) · arXiv · doi:10.48550/arxiv.2105.15203 · Cited by 5,090 · see citing works
Mobile-Former: Bridging MobileNet and Transformer
Chen, Y., Dai, X., Chen, D., Liu, M., Dong, X., Yuan, L., & Liu, Z. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.00520 · Cited by 731 · see citing works
Rep ViT: Revisiting Mobile CNN From ViT Perspective
Wang, A., Chen, H., Lin, Z., Han, J., & Ding, G. (2024) · 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52733.2024.01506 · Cited by 752 · see citing works
Rodríguez-Barroso, N., Jiménez-López, D., Luzón, M. V., Herrera, F., & Martínez-Cámara, E. (2023) · Information Fusion · doi:10.1016/j.inffus.2022.09.011 · Cited by 296 · see citing works
AdaViT: Adaptive Vision Transformers for Efficient Image Recognition
Meng, L., Li, H., Chen, B.-C., Lan, S., Wu, Z., Jiang, Y.-G., & Lim, S.-N. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01199 · Cited by 262 · see citing works
How Do Vision Transformers Work?
Park, N., & Kim, S. (2022) · arXiv · doi:10.48550/arxiv.2202.06709 · Cited by 356 · see citing works
“3 MSAs closer to the last layer significantly improve predictive performance (Graham et al., 2021; Dai et al., 2021; Xiao et al., 2021).”
BackgroundMethodResultInfluentialFQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer
Lin, Y., Zhang, T., Sun, P., Li, Z., & Zhou, S. (2022) · Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2022/164 · Cited by 198 · see citing works
“To facilitate deployment, several techniques have been proposed, including quantization [Zhou et al. , 2016; Nagel et al. , 2020; Shen et al. , 2020; Liu et al. , 2021b], pruning [Han et al. , 2016], distillation [Jiao et al. , 2020] and adaptation of architecture design [Graham et al. , 2021].”
BackgroundMethodSPViT: Enabling Faster Vision Transformers via Latency-Aware Soft Token Pruning
Kong, Z., Dong, P., Ma, X., Meng, X., Niu, W., Sun, M., Shen, X., Yuan, G., Ren, B., Tang, H., et al. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-20083-0_37 · Cited by 169 · see citing works
HRFormer: High-Resolution Transformer for Dense Prediction
Yuan, Y., Fu, R., Huang, L., Lin, W., Zhang, C., Chen, X., & Wang, J. (2021) · arXiv · doi:10.48550/arxiv.2110.09408 · Cited by 178 · see citing works
Transformers for Vision: A Survey on Innovative Methods for Computer Vision
Palanisamy, B., Hassija, V., Chatterjee, A., Mandal, A., Chakraborty, D., Pandey, A., Chalapathi, G. S. S., & Kumar, D. (2025) · IEEE Access · doi:10.1109/access.2025.3571735 · Cited by 83 · see citing works
Towards Accurate Post-Training Quantization for Vision Transformer
Ding, Y., Qin, H., Yan, Q., Chai, Z., Liu, J., Wei, X., & Liu, X. (2022) · Proceedings of the 30th ACM International Conference on Multimedia · doi:10.1145/3503161.3547826 · Cited by 79 · see citing works
Token Merging: Your ViT But Faster
Bolya, D., Fu, C.-Y., Dai, X., Zhang, P., Feichtenhofer, C., & Hoffman, J. (2022) · arXiv · doi:10.48550/arxiv.2210.09461 · Cited by 140 · see citing works
“, 2019), and some attempt to infuse domain-specific modules (Mehta & Rastegari, 2021; Graham et al., 2021; Liu et al., 2021; 2022b; Dong et al., 2022).”
BackgroundMethodInfluentialWhat Do Self-Supervised Vision Transformers Learn?
Park, N., Kim, W., Heo, B., Kim, T., & Yun, S. (2023) · arXiv · doi:10.48550/arxiv.2305.00729 · Cited by 29 · see citing works
“…features—e.g., local patterns, high-frequency signals, and texture information—and later layers capture global patterns, low-frequency signals, and shape information (Dosovitskiy et al., 2021; Raghu et al., 2021; d’Ascoli et al., 2021; Graham et al., 2021; Dai et al., 2021; Park & Kim, 2022b).”
Background
- Authors
- Graham, B., El-Nouby, A., Touvron, H., Stock, P., Joulin, A., Jegou, H., & Douze, M.
- Affiliations
- Meta (Israel) (IL)
- Université Paris Cité (FR)
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 12239-12249
- OpenAlex
- W3203166921
- Cited by
- 935
- References
- 14
- Citations by year
- 2026: 105 · 2025: 252 · 2024: 242 · 2023: 196 · 2022: 133 · 2021: 7
- FWCI
- 47.26
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
- Keywords
- vision transformer, fast inference, speed-accuracy trade-off, attentional bias, positional information, attention maps, image classification, efficient neural networks
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
