SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation
Data augmentation is vital in deep learning for enhancing model robustness by artificially expanding training datasets. However, advanced methods like CutMix blend images and assign labels based on pixel ratios, often introducing label noise by neglecting the significance of blended regions, and Sal...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10858701/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832088000634290176 |
|---|---|
| author | Hajeong Lee Zhixiong Jin Jiyoung Woo Byeongjoon Noh |
| author_facet | Hajeong Lee Zhixiong Jin Jiyoung Woo Byeongjoon Noh |
| author_sort | Hajeong Lee |
| collection | DOAJ |
| description | Data augmentation is vital in deep learning for enhancing model robustness by artificially expanding training datasets. However, advanced methods like CutMix blend images and assign labels based on pixel ratios, often introducing label noise by neglecting the significance of blended regions, and SaliencyMix applies uniform patch generation across a batch, resulting in suboptimal augmentation. This paper introduces SaliencyMix+, a novel data augmentation technique that enhances the performance of deep-learning models using saliency maps for image mixing and label generation. It identifies critical patch coordinates in batch images and refines label generation based on target object proportions, reducing label noise. Experiments on CIFAR-100 and Oxford-IIIT Pet datasets show that SaliencyMix+ consistently outperforms CutMix and SaliencyMix, achieving the lowest Top-1 errors of 24.95% and 34.89%, and Top-5 errors of 7.00% and 12.13% on CIFAR-100 and Oxford-IIIT Pet, respectively. These findings highlight the effectiveness of SaliencyMix+ in boosting model accuracy and robustness across different models and datasets. The code is publicly available on GitHub: <uri>https://github.com/SS-hj/SaliencyMixPlus.git</uri>. |
| format | Article |
| id | doaj-art-712ecd8920c14bcb954092669279526c |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-712ecd8920c14bcb954092669279526c2025-02-06T00:00:32ZengIEEEIEEE Access2169-35362025-01-0113217342174310.1109/ACCESS.2025.353708310858701SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data AugmentationHajeong Lee0https://orcid.org/0009-0007-8266-7448Zhixiong Jin1https://orcid.org/0000-0002-1370-781XJiyoung Woo2https://orcid.org/0000-0001-8231-0018Byeongjoon Noh3https://orcid.org/0000-0002-0458-1573Department of AI and Big Data, Soonchunhyang University, Asan-si, Republic of KoreaENTPE, LICIT-ECO7, Université Gustave Eiffel, Lyon, FranceDepartment of AI and Big Data, Soonchunhyang University, Asan-si, Republic of KoreaDepartment of AI and Big Data, Soonchunhyang University, Asan-si, Republic of KoreaData augmentation is vital in deep learning for enhancing model robustness by artificially expanding training datasets. However, advanced methods like CutMix blend images and assign labels based on pixel ratios, often introducing label noise by neglecting the significance of blended regions, and SaliencyMix applies uniform patch generation across a batch, resulting in suboptimal augmentation. This paper introduces SaliencyMix+, a novel data augmentation technique that enhances the performance of deep-learning models using saliency maps for image mixing and label generation. It identifies critical patch coordinates in batch images and refines label generation based on target object proportions, reducing label noise. Experiments on CIFAR-100 and Oxford-IIIT Pet datasets show that SaliencyMix+ consistently outperforms CutMix and SaliencyMix, achieving the lowest Top-1 errors of 24.95% and 34.89%, and Top-5 errors of 7.00% and 12.13% on CIFAR-100 and Oxford-IIIT Pet, respectively. These findings highlight the effectiveness of SaliencyMix+ in boosting model accuracy and robustness across different models and datasets. The code is publicly available on GitHub: <uri>https://github.com/SS-hj/SaliencyMixPlus.git</uri>.https://ieeexplore.ieee.org/document/10858701/Data augmentationSaliencyMix+saliency maplabel noise minimizationimage classification |
| spellingShingle | Hajeong Lee Zhixiong Jin Jiyoung Woo Byeongjoon Noh SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation IEEE Access Data augmentation SaliencyMix+ saliency map label noise minimization image classification |
| title | SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation |
| title_full | SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation |
| title_fullStr | SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation |
| title_full_unstemmed | SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation |
| title_short | SaliencyMix+: Noise-Minimized Image Mixing Method With Saliency Map in Data Augmentation |
| title_sort | saliencymix x002b noise minimized image mixing method with saliency map in data augmentation |
| topic | Data augmentation SaliencyMix+ saliency map label noise minimization image classification |
| url | https://ieeexplore.ieee.org/document/10858701/ |
| work_keys_str_mv | AT hajeonglee saliencymixx002bnoiseminimizedimagemixingmethodwithsaliencymapindataaugmentation AT zhixiongjin saliencymixx002bnoiseminimizedimagemixingmethodwithsaliencymapindataaugmentation AT jiyoungwoo saliencymixx002bnoiseminimizedimagemixingmethodwithsaliencymapindataaugmentation AT byeongjoonnoh saliencymixx002bnoiseminimizedimagemixingmethodwithsaliencymapindataaugmentation |