Attention transfer from human to neural networks for road object detection in winter
Abstract As an essential feature of autonomous road vehicles, obstacle detection must be executed on a real‐time onboard platform with high accuracy. Cameras are still the most commonly used sensors in autonomous driving. Most detections using cameras are based on convolutional neural networks. In t...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-11-01
|
| Series: | IET Image Processing |
| Online Access: | https://doi.org/10.1049/ipr2.12562 |
| _version_ | 1811229898140811264 |
|---|---|
| author | Jonathan Boisclair Sousso Kelouwani Follivi Kloutse Ayevide Ali Amamou Muhammad Zeshan Alam Kodjo Agbossou |
| author_facet | Jonathan Boisclair Sousso Kelouwani Follivi Kloutse Ayevide Ali Amamou Muhammad Zeshan Alam Kodjo Agbossou |
| author_sort | Jonathan Boisclair |
| collection | DOAJ |
| description | Abstract As an essential feature of autonomous road vehicles, obstacle detection must be executed on a real‐time onboard platform with high accuracy. Cameras are still the most commonly used sensors in autonomous driving. Most detections using cameras are based on convolutional neural networks. In this regard, a recent teacher–student approach, called transfer learning, has been used to improve the neural network training process. This approach has only been used with a neural network acting as a teacher to the best of our knowledge. This paper proposes a novel way of improving training data based on attention transfer by getting the attention map from a human. The proposed method allows the dataset size reduction by 50%, which leads to up to a 60% decline in the training time. The experimental results indicate that the proposed method can enhance the F1‐score of the network by up to 10% in winter conditions. |
| first_indexed | 2024-04-12T10:20:08Z |
| format | Article |
| id | doaj.art-539c6957ac104641a0d458708f33367e |
| institution | Directory Open Access Journal |
| issn | 1751-9659 1751-9667 |
| language | English |
| last_indexed | 2024-04-12T10:20:08Z |
| publishDate | 2022-11-01 |
| publisher | Wiley |
| record_format | Article |
| series | IET Image Processing |
| spelling | doaj.art-539c6957ac104641a0d458708f33367e2022-12-22T03:37:06ZengWileyIET Image Processing1751-96591751-96672022-11-0116133544355610.1049/ipr2.12562Attention transfer from human to neural networks for road object detection in winterJonathan Boisclair0Sousso Kelouwani1Follivi Kloutse Ayevide2Ali Amamou3Muhammad Zeshan Alam4Kodjo Agbossou5Hydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaHydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaHydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaHydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaHydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaHydrogen Research Institute Université du Québec á Trois‐Riviéres Trois‐Riviéres, 3351 des Forges Trois‐Riviéres QC G9A 5H7 CanadaAbstract As an essential feature of autonomous road vehicles, obstacle detection must be executed on a real‐time onboard platform with high accuracy. Cameras are still the most commonly used sensors in autonomous driving. Most detections using cameras are based on convolutional neural networks. In this regard, a recent teacher–student approach, called transfer learning, has been used to improve the neural network training process. This approach has only been used with a neural network acting as a teacher to the best of our knowledge. This paper proposes a novel way of improving training data based on attention transfer by getting the attention map from a human. The proposed method allows the dataset size reduction by 50%, which leads to up to a 60% decline in the training time. The experimental results indicate that the proposed method can enhance the F1‐score of the network by up to 10% in winter conditions.https://doi.org/10.1049/ipr2.12562 |
| spellingShingle | Jonathan Boisclair Sousso Kelouwani Follivi Kloutse Ayevide Ali Amamou Muhammad Zeshan Alam Kodjo Agbossou Attention transfer from human to neural networks for road object detection in winter IET Image Processing |
| title | Attention transfer from human to neural networks for road object detection in winter |
| title_full | Attention transfer from human to neural networks for road object detection in winter |
| title_fullStr | Attention transfer from human to neural networks for road object detection in winter |
| title_full_unstemmed | Attention transfer from human to neural networks for road object detection in winter |
| title_short | Attention transfer from human to neural networks for road object detection in winter |
| title_sort | attention transfer from human to neural networks for road object detection in winter |
| url | https://doi.org/10.1049/ipr2.12562 |
| work_keys_str_mv | AT jonathanboisclair attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter AT soussokelouwani attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter AT follivikloutseayevide attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter AT aliamamou attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter AT muhammadzeshanalam attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter AT kodjoagbossou attentiontransferfromhumantoneuralnetworksforroadobjectdetectioninwinter |