A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation
This study details the development of a lightweight and high performance model, targeting real-time object detection. Several designed features were integrated into the proposed framework to accomplish a light weight, rapid execution, and optimal performance in object detection. Foremost, a sparse a...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-02-01
|
| Series: | Electronics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9292/11/4/575 |
| _version_ | 1797480870994509824 |
|---|---|
| author | Jing-Ming Guo Jr-Sheng Yang Sankarasrinivasan Seshathiri Hung-Wei Wu |
| author_facet | Jing-Ming Guo Jr-Sheng Yang Sankarasrinivasan Seshathiri Hung-Wei Wu |
| author_sort | Jing-Ming Guo |
| collection | DOAJ |
| description | This study details the development of a lightweight and high performance model, targeting real-time object detection. Several designed features were integrated into the proposed framework to accomplish a light weight, rapid execution, and optimal performance in object detection. Foremost, a sparse and lightweight structure was chosen as the network’s backbone, and feature fusion was performed using modified feature pyramid networks. Recent learning strategies in data augmentation, mixed precision training, and network sparsity were incorporated to substantially enhance the generalization for the lightweight model and boost the detection accuracy. Moreover, knowledge distillation was applied to tackle dropping issues, and a student–teacher learning mechanism was also integrated to ensure the best performance. The model was comprehensively tested using the MS-COCO 2017 dataset, and the experimental results clearly demonstrated that the proposed model could obtain a high detection performance in comparison to state-of-the-art methods, and required minimal computational resources, making it feasible for many real-time deployments. |
| first_indexed | 2024-03-09T22:06:25Z |
| format | Article |
| id | doaj.art-48333ef993f64559a07597c3b642a9f7 |
| institution | Directory Open Access Journal |
| issn | 2079-9292 |
| language | English |
| last_indexed | 2024-03-09T22:06:25Z |
| publishDate | 2022-02-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Electronics |
| spelling | doaj.art-48333ef993f64559a07597c3b642a9f72023-11-23T19:39:34ZengMDPI AGElectronics2079-92922022-02-0111457510.3390/electronics11040575A Light-Weight CNN for Object Detection with Sparse Model and Knowledge DistillationJing-Ming Guo0Jr-Sheng Yang1Sankarasrinivasan Seshathiri2Hung-Wei Wu3Department of Electrical Engineering, National Taiwan University of Science and Technology, Taipei 106335, TaiwanDepartment of Electrical Engineering, National Taiwan University of Science and Technology, Taipei 106335, TaiwanDepartment of Electrical Engineering, National Taiwan University of Science and Technology, Taipei 106335, TaiwanDepartment of Electrical Engineering, National Taiwan University of Science and Technology, Taipei 106335, TaiwanThis study details the development of a lightweight and high performance model, targeting real-time object detection. Several designed features were integrated into the proposed framework to accomplish a light weight, rapid execution, and optimal performance in object detection. Foremost, a sparse and lightweight structure was chosen as the network’s backbone, and feature fusion was performed using modified feature pyramid networks. Recent learning strategies in data augmentation, mixed precision training, and network sparsity were incorporated to substantially enhance the generalization for the lightweight model and boost the detection accuracy. Moreover, knowledge distillation was applied to tackle dropping issues, and a student–teacher learning mechanism was also integrated to ensure the best performance. The model was comprehensively tested using the MS-COCO 2017 dataset, and the experimental results clearly demonstrated that the proposed model could obtain a high detection performance in comparison to state-of-the-art methods, and required minimal computational resources, making it feasible for many real-time deployments.https://www.mdpi.com/2079-9292/11/4/575CNNobject detectionsparse modelknowledge distillationstudent–teacher model |
| spellingShingle | Jing-Ming Guo Jr-Sheng Yang Sankarasrinivasan Seshathiri Hung-Wei Wu A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation Electronics CNN object detection sparse model knowledge distillation student–teacher model |
| title | A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation |
| title_full | A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation |
| title_fullStr | A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation |
| title_full_unstemmed | A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation |
| title_short | A Light-Weight CNN for Object Detection with Sparse Model and Knowledge Distillation |
| title_sort | light weight cnn for object detection with sparse model and knowledge distillation |
| topic | CNN object detection sparse model knowledge distillation student–teacher model |
| url | https://www.mdpi.com/2079-9292/11/4/575 |
| work_keys_str_mv | AT jingmingguo alightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT jrshengyang alightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT sankarasrinivasanseshathiri alightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT hungweiwu alightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT jingmingguo lightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT jrshengyang lightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT sankarasrinivasanseshathiri lightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation AT hungweiwu lightweightcnnforobjectdetectionwithsparsemodelandknowledgedistillation |