Learning to Drop Expensive Layers for Fast Face Recognition
Recent years have seen many advances based on Deep Convolutional Neural Networks (DCNNs) in the tasks of face recognition, most of which are developed to pursue high recognition accuracy. In this paper, we propose a novel Fast FAce Recognizer (Fast-FAR), learning to improve the speed of DCNN-based f...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2021-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9520823/ |
| _version_ | 1818402446268956672 |
|---|---|
| author | Junhui Li Wei Jia Yan Hu Shouqing Li Xiaoguang Tu |
| author_facet | Junhui Li Wei Jia Yan Hu Shouqing Li Xiaoguang Tu |
| author_sort | Junhui Li |
| collection | DOAJ |
| description | Recent years have seen many advances based on Deep Convolutional Neural Networks (DCNNs) in the tasks of face recognition, most of which are developed to pursue high recognition accuracy. In this paper, we propose a novel Fast FAce Recognizer (Fast-FAR), learning to improve the speed of DCNN-based face recognition model without sacrificing recognition accuracy. Our fundamental insight is that the computation increases exponentially with the depth of a network, the easily identifiable face images can be accurately recognized by the cheap features (pixel values at shallow layers), while the challenging samples that exhibit low quality, large pose variations or occlusions need to be processed by the expensive deep layers. The major contribution of this paper is the Reinforcement Learning Agent (RLA), which is proposed to learn a decision policy determined by a reward function. The policy adaptively decides whether the recognition should be performed at an early layer with a high recognition confidence, or proceeding to the subsequent layers, thus significantly reducing feed-forward cost for the easy faces. According to the extensive experiments on the popular face recognition benchmarks, Fast-FAR reduces the inference time by 14.22%, 20.61%, and 7.84% on the dataset LFW, AgeDB-30 and CFP-FP, respectively. |
| first_indexed | 2024-12-14T08:08:29Z |
| format | Article |
| id | doaj.art-9d563da02e5b484391cd08001d635784 |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-14T08:08:29Z |
| publishDate | 2021-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-9d563da02e5b484391cd08001d6357842022-12-21T23:10:07ZengIEEEIEEE Access2169-35362021-01-01911788011788610.1109/ACCESS.2021.31064839520823Learning to Drop Expensive Layers for Fast Face RecognitionJunhui Li0Wei Jia1Yan Hu2https://orcid.org/0000-0001-6841-0899Shouqing Li3Xiaoguang Tu4https://orcid.org/0000-0002-1185-5229Aviation Engineering Institute, Civil Aviation Flight University of China, Guanghan, ChinaSchool of Cyber Science and Engineering, Huazhong University of Science and Technology, Wuhan, ChinaAviation Engineering Institute, Civil Aviation Flight University of China, Guanghan, ChinaKey Laboratory of Flight Techniques and Flight Safety, Guanghan, CAAC, ChinaAviation Engineering Institute, Civil Aviation Flight University of China, Guanghan, ChinaRecent years have seen many advances based on Deep Convolutional Neural Networks (DCNNs) in the tasks of face recognition, most of which are developed to pursue high recognition accuracy. In this paper, we propose a novel Fast FAce Recognizer (Fast-FAR), learning to improve the speed of DCNN-based face recognition model without sacrificing recognition accuracy. Our fundamental insight is that the computation increases exponentially with the depth of a network, the easily identifiable face images can be accurately recognized by the cheap features (pixel values at shallow layers), while the challenging samples that exhibit low quality, large pose variations or occlusions need to be processed by the expensive deep layers. The major contribution of this paper is the Reinforcement Learning Agent (RLA), which is proposed to learn a decision policy determined by a reward function. The policy adaptively decides whether the recognition should be performed at an early layer with a high recognition confidence, or proceeding to the subsequent layers, thus significantly reducing feed-forward cost for the easy faces. According to the extensive experiments on the popular face recognition benchmarks, Fast-FAR reduces the inference time by 14.22%, 20.61%, and 7.84% on the dataset LFW, AgeDB-30 and CFP-FP, respectively.https://ieeexplore.ieee.org/document/9520823/Fast face recognitionreinforcement learningdeep convolutional neural networks |
| spellingShingle | Junhui Li Wei Jia Yan Hu Shouqing Li Xiaoguang Tu Learning to Drop Expensive Layers for Fast Face Recognition IEEE Access Fast face recognition reinforcement learning deep convolutional neural networks |
| title | Learning to Drop Expensive Layers for Fast Face Recognition |
| title_full | Learning to Drop Expensive Layers for Fast Face Recognition |
| title_fullStr | Learning to Drop Expensive Layers for Fast Face Recognition |
| title_full_unstemmed | Learning to Drop Expensive Layers for Fast Face Recognition |
| title_short | Learning to Drop Expensive Layers for Fast Face Recognition |
| title_sort | learning to drop expensive layers for fast face recognition |
| topic | Fast face recognition reinforcement learning deep convolutional neural networks |
| url | https://ieeexplore.ieee.org/document/9520823/ |
| work_keys_str_mv | AT junhuili learningtodropexpensivelayersforfastfacerecognition AT weijia learningtodropexpensivelayersforfastfacerecognition AT yanhu learningtodropexpensivelayersforfastfacerecognition AT shouqingli learningtodropexpensivelayersforfastfacerecognition AT xiaoguangtu learningtodropexpensivelayersforfastfacerecognition |