Rethinking Coarse-to-Fine Approach in Single Image Deblurring

By Sung-Jin Cho, Seoyoun Ji, Jun-Pyo Hong, Seung‐Won Jung, Sung-Jea Ko. IEEE International Conference on Computer Vision, 2021

862 citations recorded by the source corpus.

Coarse-to-fine strategies have been extensively used for the architecture design of single image deblurring networks. Conventional methods typically stack sub-networks with multi-scale input images and gradually improve sharpness of images from the bottom sub-network to the top sub-network, yielding inevitably high computational costs. Toward a fast and accurate deblurring network design, we revisit the coarse-to-fine strategy and present a multi-input multi-output U-net (MIMO-UNet). The MIMO-UNet has three distinct features. First, the single encoder of the MIMO-UNet takes multi-scale input images to ease the difficulty of training. Second, the single decoder of the MIMO-UNet outputs multiple deblurred images with different scales to mimic multi-cascaded U-nets using a single U-shaped network. Last, asymmetric feature fusion is introduced to merge multi-scale features in an efficient manner. Extensive experiments on the GoPro and RealBlur datasets demonstrate that the proposed network outperforms the state-of-the-art methods in terms of both accuracy and computational complexity. Source code is available for research purposes at https://github.com/chosj95/MIMO-UNet.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/ce5bf4084f63db37780dd7ccc769b352f9a3c6f7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Restormer: Efficient Transformer for High-Resolution Image Restoration (2021, methodology): “Among convolutional designs, encoder-decoder based UNet architectures [3, 14, 39, 80, 90, 93, 99] have been predominantly studied for restoration due to their hierarchical multi-scale representation while remaining computationally efficient.”
Deblurring via Stochastic Refinement (2021, methodology): “Another promising direction is to replace our initial predictor and de-noiser network with U-Net architectures that are optimized for both distortion and run time [11,14,75]. difference is size, where the starting channel count is 64 for the initial predictor and 32 for the denoiser.”
Omni-Kernel Network for Image Restoration (2024, methodology): “Our network achieves a better trade-off between performance and computation overhead over other state-of-the-art algorithms. for image restoration, such as the encoder-decoder architecture (Cho et al. 2021), residual connection (Mao et al. 2021), and attention mechanisms (Cui et al. 2023c).”
MAXIM: Multi-Axis MLP for Image Processing (2022, methodology): “ing datasets, recent high-performing approaches [14,16,19, 21, 49, 51, 59, 107, 108, 122] based on highly designed convolutional neural network (CNN) have demonstrated stateof-the-art (SOTA) performance on many tasks.”
Stripformer: Strip Transformer for Fast Image Deblurring (2022, methodology): “The deblurred results from left to right are produced by MPRNet [33], MIMO [42], DeblurGan-v2 [16], SRN [37], and our method.”
DiffIR: Efficient Diffusion Model for Image Restoration (2023, background): “Specifically, from level-1 to level-4, we set the number of dynamic transformer blocks to [1, 1, 1, 9], [13, 1, 1, 1], and [3, 5, 6, 6] for inpainting, SR, and deblurring, respectively.”
Simple Baselines for Image Restoration (2022, methodology): “Method MIMO-UNet HINet MAXIM Restormer UFormer DeepRFT MPRNet Baseline NAFNet [6] [4] [30] [37] [34] [24] -local[7] ours ours PSNR 32.”
Efficient Frequency Domain-based Transformers for High-Quality Image Deblurring (2022, methodology): “Therefore, the deblurred results by the meth-ods [3,4] still contain signiﬁcant blur effect as shown in Figure 3(c) and (g).”