Meta Transferring for Deblurring

Po-Sheng Liu (National Yang Ming Chiao Tung University),* Fu-Jen Tsai (National Tsing Hua University), Yan-Tsung Peng (National Chengchi University), Chung-Chi Tsai (Qualcomm Technology), Chia-Wen Lin (National Tsing Hua University), Yen-Yu Lin (National Yang Ming Chiao Tung University)
The 33rd British Machine Vision Conference


Previous deblurring methods devote to training a generic model with blur and sharp training pairs. However, these methods might lead to sub-optimal results caused by the domain gap between the training and testing set. In this paper, we proposed a reblur-deblur meta-transferring scheme to realize test-time adaptation for the dynamic scene deblurring. Since blur and sharp pairs are hard to obtain during testing, we leverage blurred videos to find some relative-sharp patches as pseudo ground truths, which would be reblurred by a reblurring model to form pseudo blur and sharp pairs. Our pseudo pairs can enable meta-learning to achieve test-time adaptation with few gradien updates. Extensive experimental results show that our reblur-deblur meta-learning scheme improves the existing deblurring models in various datasets, including, DVD, REDS, and RealBlur.



author    = {Po-Sheng Liu and Fu-Jen Tsai and Yan-Tsung Peng and Chung-Chi Tsai and Chia-Wen Lin and Yen-Yu Lin},
title     = {Meta Transferring for Deblurring},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year      = {2022},
url       = {}

Copyright © 2022 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection