AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring
Despite the recent progress in enhancing the efficacy of image deblurring, the limited decoding capability constrains the upper limit of State-Of-The-Art (SOTA) methods. This paper proposes a pioneering work, Adaptive Patch Exiting Reversible Decoder (AdaRevD), to explore their insufficient decoding...
Gardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Artigo |
| Publicado: |
2024
|
| Materias: | |
| Acceso en liña: | https://arxiv.org/abs/2406.09135 http://arxiv.org/abs/2406.09135 |
| Etiquetas: |
Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!
|
| _version_ | 1879175787805933568 |
|---|---|
| author | Mao, Xintian Li, Qingli Wang, Yan |
| author_browse | Li, Qingli Mao, Xintian Wang, Yan |
| author_facet | Mao, Xintian Li, Qingli Wang, Yan |
| author_sort | Mao, Xintian |
| building | arXiv |
| collection | Acceso abierto |
| description | Despite the recent progress in enhancing the efficacy of image deblurring, the limited decoding capability constrains the upper limit of State-Of-The-Art (SOTA) methods. This paper proposes a pioneering work, Adaptive Patch Exiting Reversible Decoder (AdaRevD), to explore their insufficient decoding capability. By inheriting the weights of the well-trained encoder, we refactor a reversible decoder which scales up the single-decoder training to multi-decoder training while remaining GPU memory-friendly. Meanwhile, we show that our reversible structure gradually disentangles high-level degradation degree and low-level blur pattern (residual of the blur image and its sharp counterpart) from compact degradation representation. Besides, due to the spatially-variant motion blur kernels, different blur patches have various deblurring difficulties. We further introduce a classifier to learn the degradation degree of image patches, enabling them to exit at different sub-decoders for speedup. Experiments show that our AdaRevD pushes the limit of image deblurring, e.g., achieving 34.60 dB in PSNR on GoPro dataset. |
| first_indexed | 2026-10-16T00:34:38Z |
| format | Article |
| fullrecord | <oai_dc:dc xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ https://www.openarchives.org/OAI/2.0/oai_dc.xsd"><identifier>arxiv-2406.09135</identifier><datestamp>2024-06-14</datestamp>
<dc:title>AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring</dc:title>
<dc:creator>Mao, Xintian</dc:creator>
<dc:creator>Li, Qingli</dc:creator>
<dc:creator>Wang, Yan</dc:creator>
<dc:subject>Computer Vision and Pattern Recognition</dc:subject>
<dc:description>Despite the recent progress in enhancing the efficacy of image deblurring, the limited decoding capability constrains the upper limit of State-Of-The-Art (SOTA) methods. This paper proposes a pioneering work, Adaptive Patch Exiting Reversible Decoder (AdaRevD), to explore their insufficient decoding capability. By inheriting the weights of the well-trained encoder, we refactor a reversible decoder which scales up the single-decoder training to multi-decoder training while remaining GPU memory-friendly. Meanwhile, we show that our reversible structure gradually disentangles high-level degradation degree and low-level blur pattern (residual of the blur image and its sharp counterpart) from compact degradation representation. Besides, due to the spatially-variant motion blur kernels, different blur patches have various deblurring difficulties. We further introduce a classifier to learn the degradation degree of image patches, enabling them to exit at different sub-decoders for speedup. Experiments show that our AdaRevD pushes the limit of image deblurring, e.g., achieving 34.60 dB in PSNR on GoPro dataset. </dc:description>
<dc:date>2024-06-13</dc:date>
<dc:type>text</dc:type>
<dc:identifier>https://arxiv.org/abs/2406.09135</dc:identifier>
<dc:identifier>http://arxiv.org/abs/2406.09135</dc:identifier>
</oai_dc:dc> |
| id | arxiv-2406.09135 |
| institution | arXiv |
| last_indexed | 2026-10-16T00:34:38Z |
| publishDate | 2024 |
| publishDateRange | 2024 |
| publishDateSort | 2024 |
| record_format | oai_dc |
| spelling | arxiv-2406.091352024-06-14 AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring Mao, Xintian Li, Qingli Wang, Yan Computer Vision and Pattern Recognition Despite the recent progress in enhancing the efficacy of image deblurring, the limited decoding capability constrains the upper limit of State-Of-The-Art (SOTA) methods. This paper proposes a pioneering work, Adaptive Patch Exiting Reversible Decoder (AdaRevD), to explore their insufficient decoding capability. By inheriting the weights of the well-trained encoder, we refactor a reversible decoder which scales up the single-decoder training to multi-decoder training while remaining GPU memory-friendly. Meanwhile, we show that our reversible structure gradually disentangles high-level degradation degree and low-level blur pattern (residual of the blur image and its sharp counterpart) from compact degradation representation. Besides, due to the spatially-variant motion blur kernels, different blur patches have various deblurring difficulties. We further introduce a classifier to learn the degradation degree of image patches, enabling them to exit at different sub-decoders for speedup. Experiments show that our AdaRevD pushes the limit of image deblurring, e.g., achieving 34.60 dB in PSNR on GoPro dataset. 2024-06-13 text https://arxiv.org/abs/2406.09135 http://arxiv.org/abs/2406.09135 |
| spellingShingle | Computer Vision and Pattern Recognition Mao, Xintian Li, Qingli Wang, Yan AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title_full | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title_fullStr | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title_full_unstemmed | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title_short | AdaRevD: Adaptive Patch Exiting Reversible Decoder Pushes the Limit of Image Deblurring |
| title_sort | adarevd adaptive patch exiting reversible decoder pushes the limit of image deblurring |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2406.09135 http://arxiv.org/abs/2406.09135 |
| work_keys_str_mv | AT maoxintian adarevdadaptivepatchexitingreversibledecoderpushesthelimitofimagedeblurring AT liqingli adarevdadaptivepatchexitingreversibledecoderpushesthelimitofimagedeblurring AT wangyan adarevdadaptivepatchexitingreversibledecoderpushesthelimitofimagedeblurring |