The spatially-correlative loss for various image translation tasks
We propose a novel spatially-correlative loss that is simple, efficient and yet effective for preserving scene structure consistency while supporting large appearance changes during unpaired image-to-image (I2I) translation. Previous methods attempt this by using pixel-level cycle-consistency or fea...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Conference Paper |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/151225 |
| _version_ | 1826123097497927680 |
|---|---|
| author | Zheng, Chuanxia Cham, Tat-Jen Cai, Jianfei |
| author2 | School of Computer Science and Engineering |
| author_facet | School of Computer Science and Engineering Zheng, Chuanxia Cham, Tat-Jen Cai, Jianfei |
| author_sort | Zheng, Chuanxia |
| collection | NTU |
| description | We propose a novel spatially-correlative loss that is simple, efficient and yet effective for preserving scene structure consistency while supporting large appearance changes during unpaired image-to-image (I2I) translation. Previous methods attempt this by using pixel-level cycle-consistency or feature-level matching losses, but the domain-specific nature of these losses hinder translation across large domain gaps. To address this, we exploit the spatial patterns of self-similarity as a means of defining scene structure. Our spatially-correlative loss is geared towards only capturing spatial relationships within an image rather than domain appearance. We also introduce a new self-supervised learning method to explicitly learn spatially-correlative maps for each specific translation task. We show distinct improvement over baseline models in all three modes of unpaired I2I translation: single-modal, multi-modal, and even single-image translation. This new loss can easily be integrated into existing network architectures and thus allows wide applicability. |
| first_indexed | 2024-10-01T05:59:04Z |
| format | Conference Paper |
| id | ntu-10356/151225 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T05:59:04Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | ntu-10356/1512252021-06-11T11:14:41Z The spatially-correlative loss for various image translation tasks Zheng, Chuanxia Cham, Tat-Jen Cai, Jianfei School of Computer Science and Engineering IEEE Conference on Computer Vision and Pattern Recognition Engineering::Computer science and engineering::Computing methodologies::Pattern recognition Spatially-correlative Loss Translation Tasks We propose a novel spatially-correlative loss that is simple, efficient and yet effective for preserving scene structure consistency while supporting large appearance changes during unpaired image-to-image (I2I) translation. Previous methods attempt this by using pixel-level cycle-consistency or feature-level matching losses, but the domain-specific nature of these losses hinder translation across large domain gaps. To address this, we exploit the spatial patterns of self-similarity as a means of defining scene structure. Our spatially-correlative loss is geared towards only capturing spatial relationships within an image rather than domain appearance. We also introduce a new self-supervised learning method to explicitly learn spatially-correlative maps for each specific translation task. We show distinct improvement over baseline models in all three modes of unpaired I2I translation: single-modal, multi-modal, and even single-image translation. This new loss can easily be integrated into existing network architectures and thus allows wide applicability. 2021-06-11T11:14:40Z 2021-06-11T11:14:40Z 2021 Conference Paper Zheng, C., Cham, T. & Cai, J. (2021). The spatially-correlative loss for various image translation tasks. IEEE Conference on Computer Vision and Pattern Recognition. https://hdl.handle.net/10356/151225 en © 2021 Institute of Electrical and Electronics Engineers (IEEE). All rights reserved. |
| spellingShingle | Engineering::Computer science and engineering::Computing methodologies::Pattern recognition Spatially-correlative Loss Translation Tasks Zheng, Chuanxia Cham, Tat-Jen Cai, Jianfei The spatially-correlative loss for various image translation tasks |
| title | The spatially-correlative loss for various image translation tasks |
| title_full | The spatially-correlative loss for various image translation tasks |
| title_fullStr | The spatially-correlative loss for various image translation tasks |
| title_full_unstemmed | The spatially-correlative loss for various image translation tasks |
| title_short | The spatially-correlative loss for various image translation tasks |
| title_sort | spatially correlative loss for various image translation tasks |
| topic | Engineering::Computer science and engineering::Computing methodologies::Pattern recognition Spatially-correlative Loss Translation Tasks |
| url | https://hdl.handle.net/10356/151225 |
| work_keys_str_mv | AT zhengchuanxia thespatiallycorrelativelossforvariousimagetranslationtasks AT chamtatjen thespatiallycorrelativelossforvariousimagetranslationtasks AT caijianfei thespatiallycorrelativelossforvariousimagetranslationtasks AT zhengchuanxia spatiallycorrelativelossforvariousimagetranslationtasks AT chamtatjen spatiallycorrelativelossforvariousimagetranslationtasks AT caijianfei spatiallycorrelativelossforvariousimagetranslationtasks |