Rain to Rain: Learning Real Rain Removal Without Ground Truth
Image deraining is a low-level restoration task that has become quite popular during the past decades. Although recent data-driven deraining models exhibit promising results, most of these models are trained on synthetic rain data sets which do not generalize well when applied to real rain images. W...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2021-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9400826/ |
| _version_ | 1818956054133735424 |
|---|---|
| author | Abderraouf Khodja Zhonglong Zheng Jiashuaizi Mo Dawei Zhang Liyuan Chen |
| author_facet | Abderraouf Khodja Zhonglong Zheng Jiashuaizi Mo Dawei Zhang Liyuan Chen |
| author_sort | Abderraouf Khodja |
| collection | DOAJ |
| description | Image deraining is a low-level restoration task that has become quite popular during the past decades. Although recent data-driven deraining models exhibit promising results, most of these models are trained on synthetic rain data sets which do not generalize well when applied to real rain images. While recent real-rain data sets have achieved favorable generalization performance, generating rain-free ground-truths can be tedious and time-consuming. To address this problem, in this work, we present rain to rain training, an unsupervised training method for single image deraining. Our experiments show that it is possible to train single image deraining models by using only rain images. This can be achieved by simply training models to map pairs of rain images. We also introduce the idea of using the least overlapping training pairs, a method of selecting adequate training pairs that enables rain to rain training to achieve equivalent deraining performance compared to supervised training. |
| first_indexed | 2024-12-20T10:47:51Z |
| format | Article |
| id | doaj.art-db778bf8e7f94c34bbaf2bda25459dfc |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-20T10:47:51Z |
| publishDate | 2021-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-db778bf8e7f94c34bbaf2bda25459dfc2022-12-21T19:43:20ZengIEEEIEEE Access2169-35362021-01-019573255733710.1109/ACCESS.2021.30726879400826Rain to Rain: Learning Real Rain Removal Without Ground TruthAbderraouf Khodja0https://orcid.org/0000-0003-0267-8117Zhonglong Zheng1https://orcid.org/0000-0002-5271-9215Jiashuaizi Mo2https://orcid.org/0000-0001-5801-0795Dawei Zhang3https://orcid.org/0000-0002-7593-1593Liyuan Chen4College of Mathematics and Computer Science, Zhejiang Normal University, Jinhua, ChinaCollege of Mathematics and Computer Science, Zhejiang Normal University, Jinhua, ChinaCollege of Mathematics and Computer Science, Zhejiang Normal University, Jinhua, ChinaCollege of Mathematics and Computer Science, Zhejiang Normal University, Jinhua, ChinaCollege of Mathematics and Computer Science, Zhejiang Normal University, Jinhua, ChinaImage deraining is a low-level restoration task that has become quite popular during the past decades. Although recent data-driven deraining models exhibit promising results, most of these models are trained on synthetic rain data sets which do not generalize well when applied to real rain images. While recent real-rain data sets have achieved favorable generalization performance, generating rain-free ground-truths can be tedious and time-consuming. To address this problem, in this work, we present rain to rain training, an unsupervised training method for single image deraining. Our experiments show that it is possible to train single image deraining models by using only rain images. This can be achieved by simply training models to map pairs of rain images. We also introduce the idea of using the least overlapping training pairs, a method of selecting adequate training pairs that enables rain to rain training to achieve equivalent deraining performance compared to supervised training.https://ieeexplore.ieee.org/document/9400826/Image restorationreal rainsynthetic rainsingle image derainingunsupervised training |
| spellingShingle | Abderraouf Khodja Zhonglong Zheng Jiashuaizi Mo Dawei Zhang Liyuan Chen Rain to Rain: Learning Real Rain Removal Without Ground Truth IEEE Access Image restoration real rain synthetic rain single image deraining unsupervised training |
| title | Rain to Rain: Learning Real Rain Removal Without Ground Truth |
| title_full | Rain to Rain: Learning Real Rain Removal Without Ground Truth |
| title_fullStr | Rain to Rain: Learning Real Rain Removal Without Ground Truth |
| title_full_unstemmed | Rain to Rain: Learning Real Rain Removal Without Ground Truth |
| title_short | Rain to Rain: Learning Real Rain Removal Without Ground Truth |
| title_sort | rain to rain learning real rain removal without ground truth |
| topic | Image restoration real rain synthetic rain single image deraining unsupervised training |
| url | https://ieeexplore.ieee.org/document/9400826/ |
| work_keys_str_mv | AT abderraoufkhodja raintorainlearningrealrainremovalwithoutgroundtruth AT zhonglongzheng raintorainlearningrealrainremovalwithoutgroundtruth AT jiashuaizimo raintorainlearningrealrainremovalwithoutgroundtruth AT daweizhang raintorainlearningrealrainremovalwithoutgroundtruth AT liyuanchen raintorainlearningrealrainremovalwithoutgroundtruth |