Timezone: »
Convolutions are a fundamental building block of modern computer vision systems. Recent approaches have argued for going beyond convolutions in order to capture long-range dependencies. These efforts focus on augmenting convolutional models with content-based interactions, such as self-attention and non-local means, to achieve gains on a number of vision tasks. The natural question that arises is whether attention can be a stand-alone primitive for vision models instead of serving as just an augmentation on top of convolutions. In developing and testing a pure self-attention vision model, we verify that self-attention can indeed be an effective stand-alone layer. A simple procedure of replacing all instances of spatial convolutions with a form of self-attention to ResNet-50 produces a fully self-attentional model that outperforms the baseline on ImageNet classification with 12% fewer FLOPS and 29% fewer parameters. On COCO object detection, a fully self-attention model matches the mAP of a baseline RetinaNet while having 39% fewer FLOPS and 34% fewer parameters. Detailed ablation studies demonstrate that self-attention is especially impactful when used in later layers. These results establish that stand-alone self-attention is an important addition to the vision practitioner's toolbox.
Author Information
Niki Parmar (Google)
Prajit Ramachandran (Google Brain)
Ashish Vaswani (Google Brain)
Irwan Bello (Google Brain)
Anselm Levskaya (Google)
Jonathon Shlens (Google Research)
More from the Same Authors
-
2021 Spotlight: Revisiting ResNets: Improved Training and Scaling Strategies »
Irwan Bello · William Fedus · Xianzhi Du · Ekin Dogus Cubuk · Aravind Srinivas · Tsung-Yi Lin · Jonathon Shlens · Barret Zoph -
2022 Workshop: Vision Transformers: Theory and applications »
Fahad Shahbaz Khan · Gul Varol · Salman Khan · Ping Luo · Rao Anwer · Ashish Vaswani · Hisham Cholakkal · Niki Parmar · Joost van de Weijer · Mubarak Shah -
2021 Poster: Revisiting ResNets: Improved Training and Scaling Strategies »
Irwan Bello · William Fedus · Xianzhi Du · Ekin Dogus Cubuk · Aravind Srinivas · Tsung-Yi Lin · Jonathon Shlens · Barret Zoph -
2020 Poster: RandAugment: Practical Automated Data Augmentation with a Reduced Search Space »
Ekin Dogus Cubuk · Barret Zoph · Jonathon Shlens · Quoc V Le -
2019 : Coffee Break + Poster Session II »
Niki Parmar · Haraldur Hallgrimsson · Christian Kames · Arijit Patra · Abdullah-Al-Zubaer Imran · Junlin Yang · David Zimmerer · Arunava Chakravarty · Lawrence Schobs · Alexej Gossmann · TUNG-I CHEN · Tarun Dutt · Li Yao · Octavio Eleazar Martinez Manzanera · Johannes Pinckaers · Mehmet Ufuk Dalmis · Deepak Gupta · Nandinee Haq · David Ruhe · Jevgenij Gamper · Alfredo De Goyeneche Macaya · Jonathan Tamir · Byunghwan Jeon · SUBBAREDDY OOTA · Reinhard Heckel · Pamela K Douglas · Oleksii Sidorov · Ke Wang · Melanie Garcia · Ravi Soni · Ankita Shukla -
2019 : Oral Session III – Imaging »
Niki Parmar · Haraldur Hallgrimsson · Christian Kames -
2019 Poster: A Fourier Perspective on Model Robustness in Computer Vision »
Dong Yin · Raphael Gontijo Lopes · Jonathon Shlens · Ekin Dogus Cubuk · Justin Gilmer -
2018 Poster: Searching for Efficient Multi-Scale Architectures for Dense Image Prediction »
Liang-Chieh Chen · Maxwell Collins · Yukun Zhu · George Papandreou · Barret Zoph · Florian Schroff · Hartwig Adam · Jonathon Shlens -
2018 Poster: Mesh-TensorFlow: Deep Learning for Supercomputers »
Noam Shazeer · Youlong Cheng · Niki Parmar · Dustin Tran · Ashish Vaswani · Penporn Koanantakool · Peter Hawkins · HyoukJoong Lee · Mingsheng Hong · Cliff Young · Ryan Sepassi · Blake Hechtman -
2017 Poster: Attention is All you Need »
Ashish Vaswani · Noam Shazeer · Niki Parmar · Jakob Uszkoreit · Llion Jones · Aidan Gomez · Łukasz Kaiser · Illia Polosukhin -
2017 Spotlight: Attention is All you Need »
Ashish Vaswani · Noam Shazeer · Niki Parmar · Jakob Uszkoreit · Llion Jones · Aidan Gomez · Łukasz Kaiser · Illia Polosukhin -
2013 Poster: DeViSE: A Deep Visual-Semantic Embedding Model »
Andrea Frome · Greg Corrado · Jonathon Shlens · Samy Bengio · Jeff Dean · Marc'Aurelio Ranzato · Tomas Mikolov -
2013 Demonstration: DeViSE: A Deep Visual-Semantic Embedding Model »
Jonathon Shlens · Andrea Frome