Realistic Blur Synthesis for Learning Image Deblurring

Rim, Jaesung; Kim, Geonung; Kim, Jungeon; Liu, Junyong; Lee, Seungyong; Cho, Sunghyun

doi:10.1007/978-3-031-20071-7_29

Cited by 34 publications

(11 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To alleviate this problem, researchers either use frame interpolation technology to increase the frame rate [13] or capture videos at a high frame rate (e.g., 1000 fps) [14]. Recently, the works [18,23] synthesized realistic blur data using handcrafted ISP that is based on a specific camera imaging system. However, existing synthetic datasets do not generalize well to real, blurred images.…”

Section: Image Deblurring Datasetsmentioning

confidence: 99%

“…Since the CRF is difficult to estimate accurately, it is not ideal to measure it from multi-exposure scene shots [13] or to approximate it as the gamma function [6]. Recently, some works [18,23] synthesize blur in the RAW domain and then use handcrafted ISP to convert the RAW image to an RGB image. However, the handcrafted ISP affects the generalization ability of the deblurring models to handle real-world blurs captured by other hand-held devices [31].…”

Section: Preliminariesmentioning

confidence: 99%

“…To achieve this, we use the pre-trained CycleISP [30], which takes an RGB image as input and produces a corresponding synthesized RAW image. We use the reverse neural ISP instead of handcrafted, which is different from previous blur synthesis methods [18,23].…”

Section: Raw Image Reconstructionmentioning

confidence: 99%

“…We adopt the widely used GoPro [6], RSBlur [18], and the proposed synthetic dataset in the experiments. There are 2103 training images with dynamic motion blur in the GoPro dataset and 8878 training images in the RSBlur dataset.…”

Section: Training Datasetsmentioning

confidence: 99%

“…We evaluate the proposed method on the synthetic datasets (GoPro [6] and HIDE [21]) and the real datasets (BSD [16], RSBlur-Real [18], RealBlur-J [15], and Lai [17]). The GoPro dataset contains 1,111 pairs of blurred and sharp images for testing.…”

Section: Testing Datasetsmentioning

confidence: 99%

See 4 more Smart Citations

Real‐world image deblurring using data synthesis and feature complementary network

Wei,

Ge,

Qiao

et al. 2024

IET Image Processing

View full text Add to dashboard Cite

Many learning‐based approaches to image deblurring have received increasing attention in recent years. However, the models trained on existing synthetic datasets do not generalize well to real‐world blur, resulting in undesirable artifacts and residual blur. This work attempts to address this problem from two aspects: training data synthesis and network architecture. To narrow the domain gap between synthetic and real domains, a realistic blur synthesis pipeline to generate high‐quality blurred data is proposed. Since the blur is non‐uniform and has different scales and degrees, a parallel feature complementary module to fully exploit the local and non‐local information, which improves the feature representation and helps the network to perceive the non‐uniform blur, is developed. In addition, a spatial Fourier reconstruction block to facilitate correct detail recovery in the spatial and Fourier domains is introduced. Based on these two designs, an effective encoder–decoder network for deblurring is designed. Extensive experiments demonstrate the validity and superiority of the proposed blur synthesis method and deblurring network. In particular, the proposed deblurring network can achieve superior or comparable performance to Restormer, while saving 70% of network parameters and 53% of floating point operations (FLOPs).

show abstract