The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Model pengesanan objek berasaskan pembelajaran mendalam berprestasi tinggi boleh mengurangkan kemalangan lalu lintas menggunakan imej kamera pemuka semasa memandu waktu malam. Pembelajaran mendalam memerlukan set data berskala besar untuk mendapatkan model berprestasi tinggi. Walau bagaimanapun, set data pengesanan objek sedia ada kebanyakannya adalah adegan siang hari dan beberapa adegan waktu malam. Meningkatkan set data waktu malam adalah sukar dan memakan masa. Dalam kes sedemikian, adalah mungkin untuk menukar imej siang hari kepada imej waktu malam dengan model terjemahan imej-ke-imej untuk menambah set data waktu malam dengan sedikit usaha supaya set data yang diterjemahkan boleh menggunakan anotasi set data siang hari. Oleh itu, dalam kajian ini, model terjemahan imej-ke-imej berasaskan GAN dicadangkan dengan menggabungkan perhatian kendiri dengan ketekalan kitaran dan pemisahan kandungan/gaya untuk penambahan data waktu malam yang menunjukkan kesetiaan tinggi kepada anotasi set data siang hari. Keputusan eksperimen menyerlahkan keberkesanan model yang dicadangkan berbanding dengan model lain dari segi imej terjemahan dan skor FID. Selain itu, kesetiaan tinggi imej yang diterjemahkan kepada anotasi disahkan oleh model pengesanan objek kecil mengikut hasil pengesanan dan mAP. Kajian ablasi mengesahkan keberkesanan perhatian diri dalam model yang dicadangkan. Sebagai sumbangan kepada penambahan data berasaskan GAN, kod sumber model terjemahan imej yang dicadangkan tersedia secara terbuka di https://github.com/subecky/Image-Translation-With-Self-Attention
Rebeka SULTANA
Shizuoka University
Gosuke OHASHI
Shizuoka University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Salinan
Rebeka SULTANA, Gosuke OHASHI, "GAN-based Image Translation Model with Self-Attention for Nighttime Dashcam Data Augmentation" in IEICE TRANSACTIONS on Fundamentals,
vol. E106-A, no. 9, pp. 1202-1210, September 2023, doi: 10.1587/transfun.2022IMP0004.
Abstract: High-performance deep learning-based object detection models can reduce traffic accidents using dashcam images during nighttime driving. Deep learning requires a large-scale dataset to obtain a high-performance model. However, existing object detection datasets are mostly daytime scenes and a few nighttime scenes. Increasing the nighttime dataset is laborious and time-consuming. In such a case, it is possible to convert daytime images to nighttime images by image-to-image translation model to augment the nighttime dataset with less effort so that the translated dataset can utilize the annotations of the daytime dataset. Therefore, in this study, a GAN-based image-to-image translation model is proposed by incorporating self-attention with cycle consistency and content/style separation for nighttime data augmentation that shows high fidelity to annotations of the daytime dataset. Experimental results highlight the effectiveness of the proposed model compared with other models in terms of translated images and FID scores. Moreover, the high fidelity of translated images to the annotations is verified by a small object detection model according to detection results and mAP. Ablation studies confirm the effectiveness of self-attention in the proposed model. As a contribution to GAN-based data augmentation, the source code of the proposed image translation model is publicly available at https://github.com/subecky/Image-Translation-With-Self-Attention
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.2022IMP0004/_p
Salinan
@ARTICLE{e106-a_9_1202,
author={Rebeka SULTANA, Gosuke OHASHI, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={GAN-based Image Translation Model with Self-Attention for Nighttime Dashcam Data Augmentation},
year={2023},
volume={E106-A},
number={9},
pages={1202-1210},
abstract={High-performance deep learning-based object detection models can reduce traffic accidents using dashcam images during nighttime driving. Deep learning requires a large-scale dataset to obtain a high-performance model. However, existing object detection datasets are mostly daytime scenes and a few nighttime scenes. Increasing the nighttime dataset is laborious and time-consuming. In such a case, it is possible to convert daytime images to nighttime images by image-to-image translation model to augment the nighttime dataset with less effort so that the translated dataset can utilize the annotations of the daytime dataset. Therefore, in this study, a GAN-based image-to-image translation model is proposed by incorporating self-attention with cycle consistency and content/style separation for nighttime data augmentation that shows high fidelity to annotations of the daytime dataset. Experimental results highlight the effectiveness of the proposed model compared with other models in terms of translated images and FID scores. Moreover, the high fidelity of translated images to the annotations is verified by a small object detection model according to detection results and mAP. Ablation studies confirm the effectiveness of self-attention in the proposed model. As a contribution to GAN-based data augmentation, the source code of the proposed image translation model is publicly available at https://github.com/subecky/Image-Translation-With-Self-Attention},
keywords={},
doi={10.1587/transfun.2022IMP0004},
ISSN={1745-1337},
month={September},}
Salinan
TY - JOUR
TI - GAN-based Image Translation Model with Self-Attention for Nighttime Dashcam Data Augmentation
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1202
EP - 1210
AU - Rebeka SULTANA
AU - Gosuke OHASHI
PY - 2023
DO - 10.1587/transfun.2022IMP0004
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E106-A
IS - 9
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - September 2023
AB - High-performance deep learning-based object detection models can reduce traffic accidents using dashcam images during nighttime driving. Deep learning requires a large-scale dataset to obtain a high-performance model. However, existing object detection datasets are mostly daytime scenes and a few nighttime scenes. Increasing the nighttime dataset is laborious and time-consuming. In such a case, it is possible to convert daytime images to nighttime images by image-to-image translation model to augment the nighttime dataset with less effort so that the translated dataset can utilize the annotations of the daytime dataset. Therefore, in this study, a GAN-based image-to-image translation model is proposed by incorporating self-attention with cycle consistency and content/style separation for nighttime data augmentation that shows high fidelity to annotations of the daytime dataset. Experimental results highlight the effectiveness of the proposed model compared with other models in terms of translated images and FID scores. Moreover, the high fidelity of translated images to the annotations is verified by a small object detection model according to detection results and mAP. Ablation studies confirm the effectiveness of self-attention in the proposed model. As a contribution to GAN-based data augmentation, the source code of the proposed image translation model is publicly available at https://github.com/subecky/Image-Translation-With-Self-Attention
ER -