Timezone: »
Recently, Vision Transformer and its variants have shown great promise on various computer vision tasks. The ability to capture local and global visual dependencies through self-attention is the key to its success. But it also brings challenges due to quadratic computational overhead, especially for the high-resolution vision tasks(e.g., object detection). Many recent works have attempted to reduce the cost and improve model performance by applying either coarse-grained global attention or fine-grained local attention. However, both approaches cripple the modeling power of the original self-attention mechanism of multi-layer Transformers, leading to sub-optimal solutions. In this paper, we present focal attention, a new attention mechanism that incorporates both fine-grained local and coarse-grained global interactions. In this new mechanism, each token attends its closest surrounding tokens at the fine granularity and the tokens far away at a coarse granularity and thus can capture both short- and long-range visual dependencies efficiently and effectively. With focal attention, we propose a new variant of Vision Transformer models, called Focal Transformers, which achieve superior performance over the state-of-the-art (SoTA) Vision Transformers on a range of public image classification and object detection benchmarks. In particular, our Focal Transformer models with a moderate size of 51.1M and a large size of 89.8M achieve 83.6% and 84.0%Top-1 accuracy, respectively, on ImageNet classification at 224×224. When employed as the backbones, Focal Transformers achieve consistent and substantial improvements over the current SoTA Swin Transformers [44] across 6 different object detection methods. Our largest Focal Transformer yields58.7/59.0boxmAPs and50.9/51.3mask mAPs on COCO mini-val/test-dev, and55.4mIoU onADE20K for semantic segmentation, creating new SoTA on three of the most challenging computer vision tasks.
Author Information
Jianwei Yang (Microsoft Research)
Chunyuan Li (Duke University)
Pengchuan Zhang (Microsoft Research)
Xiyang Dai (Microsoft)
Bin Xiao (South China University of Technology)
Lu Yuan (Microsoft)
Jianfeng Gao (Microsoft Research, Redmond, WA)
Related Events (a corresponding poster, oral, or spotlight)
-
2021 Poster: Focal Attention for Long-Range Interactions in Vision Transformers »
Thu. Dec 9th 12:30 -- 02:00 AM Room None
More from the Same Authors
-
2021 : Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models »
Boxin Wang · Chejian Xu · Shuohang Wang · Zhe Gan · Yu Cheng · Jianfeng Gao · Ahmed Awadallah · Bo Li -
2021 : Few-Shot Learning Evaluation in Natural Language Understanding »
Subhabrata Mukherjee · Xiaodong Liu · Guoqing Zheng · Saghar Hosseini · Hao Cheng · Ge Yang · Christopher Meek · Ahmed Awadallah · Jianfeng Gao -
2021 : Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models »
Boxin Wang · Chejian Xu · Shuohang Wang · Zhe Gan · Yu Cheng · Jianfeng Gao · Ahmed Awadallah · Bo Li -
2021 Poster: Stronger NAS with Weaker Predictors »
Junru Wu · Xiyang Dai · Dongdong Chen · Yinpeng Chen · Mengchen Liu · Ye Yu · Zhangyang Wang · Zicheng Liu · Mei Chen · Lu Yuan -
2021 Poster: Chasing Sparsity in Vision Transformers: An End-to-End Exploration »
Tianlong Chen · Yu Cheng · Zhe Gan · Lu Yuan · Lei Zhang · Zhangyang Wang -
2021 Poster: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer »
Ge Yang · Edward Hu · Igor Babuschkin · Szymon Sidor · Xiaodong Liu · David Farhi · Nick Ryder · Jakub Pachocki · Weizhu Chen · Jianfeng Gao -
2021 : WebQA Competition + Q&A »
Yingshan CHANG · Yonatan Bisk · Mridu Narang · Levi Melnick · Jianfeng Gao · Hisami Suzuki · Guihong Cao -
2020 Poster: GreedyFool: Distortion-Aware Sparse Adversarial Attack »
Xiaoyi Dong · Dongdong Chen · Jianmin Bao · Chuan Qin · Lu Yuan · Weiming Zhang · Nenghai Yu · Dong Chen -
2019 Poster: A Convex Relaxation Barrier to Tight Robustness Verification of Neural Networks »
Hadi Salman · Greg Yang · Huan Zhang · Cho-Jui Hsieh · Pengchuan Zhang -
2019 Poster: Unified Language Model Pre-training for Natural Language Understanding and Generation »
Li Dong · Nan Yang · Wenhui Wang · Furu Wei · Xiaodong Liu · Yu Wang · Jianfeng Gao · Ming Zhou · Hsiao-Wuen Hon -
2019 Poster: Using Statistics to Automate Stochastic Optimization »
Hunter Lang · Lin Xiao · Pengchuan Zhang -
2019 Poster: Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers »
Hadi Salman · Jerry Li · Ilya Razenshteyn · Pengchuan Zhang · Huan Zhang · Sebastien Bubeck · Greg Yang -
2019 Spotlight: Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers »
Hadi Salman · Jerry Li · Ilya Razenshteyn · Pengchuan Zhang · Huan Zhang · Sebastien Bubeck · Greg Yang -
2019 Poster: Understanding the Role of Momentum in Stochastic Gradient Methods »
Igor Gitman · Hunter Lang · Pengchuan Zhang · Lin Xiao -
2018 Poster: M-Walk: Learning to Walk over Graphs using Monte Carlo Tree Search »
Yelong Shen · Jianshu Chen · Po-Sen Huang · Yuqing Guo · Jianfeng Gao -
2018 Poster: Generating Informative and Diverse Conversational Responses via Adversarial Information Maximization »
Yizhe Zhang · Michel Galley · Jianfeng Gao · Zhe Gan · Xiujun Li · Chris Brockett · Bill Dolan -
2018 Poster: Turbo Learning for CaptionBot and DrawingBot »
Qiuyuan Huang · Pengchuan Zhang · Dapeng Wu · Lei Zhang -
2018 Poster: Navigating with Graph Representations for Fast and Scalable Decoding of Neural Language Models »
Minjia Zhang · Wenhan Wang · Xiaodong Liu · Jianfeng Gao · Yuxiong He -
2017 : Invited Talk: Microsoft (Asli and Jianfeng) »
Jianfeng Gao -
2017 Poster: Triangle Generative Adversarial Networks »
Zhe Gan · Liqun Chen · Weiyao Wang · Yuchen Pu · Yizhe Zhang · Hao Liu · Chunyuan Li · Lawrence Carin -
2015 Poster: End-to-end Learning of LDA by Mirror-Descent Back Propagation over a Deep Architecture »
Jianshu Chen · Ji He · Yelong Shen · Lin Xiao · Xiaodong He · Jianfeng Gao · Xinying Song · Li Deng -
2015 Poster: Deep Temporal Sigmoid Belief Networks for Sequence Modeling »
Zhe Gan · Chunyuan Li · Ricardo Henao · David Carlson · Lawrence Carin