Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deblurring in the Wild: A Real-World Image Deblurring Dataset from Smartphone High-Speed Videos

Syed Mumtahin Mahmud, Mahdi Mohd Hossain Noki, Prothito Shovon Majumder, Abdul Mohaimen Al Radi, Sudipto Das Sukanto, Afia Lubaina, Md. Mosaddek Khan

TL;DR

This work addresses the gap in real-world image deblurring benchmarks by introducing a large-scale, smartphone-based dataset built from 240 fps slow-motion video to synthesize realistic long-exposure blur. The authors generate over 42,000 blur–sharp pairs at 1920×1080 by averaging 30 consecutive frames, with the temporally centered frame serving as ground truth, yielding an effective exposure of $T = 1/8$ s. They benchmark several state-of-the-art deblurring models and observe significant performance degradation on this realistic dataset, underscoring the challenge of non-uniform, real-world blur and the need for more robust models. The dataset, captured on an iPhone 15 Pro, offers large scale, diversity, and domain relevance for training and evaluating deblurring methods aimed at consumer photography, paving the way for better generalization to practical smartphone imagery. The work highlights the real-world relevance of frame-averaging blur synthesis and motivates future research into more perceptually aligned and hardware-aware deblurring approaches.

Abstract

We introduce the largest real-world image deblurring dataset constructed from smartphone slow-motion videos. Using 240 frames captured over one second, we simulate realistic long-exposure blur by averaging frames to produce blurry images, while using the temporally centered frame as the sharp reference. Our dataset contains over 42,000 high-resolution blur-sharp image pairs, making it approximately 10 times larger than widely used datasets, with 8 times the amount of different scenes, including indoor and outdoor environments, with varying object and camera motions. We benchmark multiple state-of-the-art (SOTA) deblurring models on our dataset and observe significant performance degradation, highlighting the complexity and diversity of our benchmark. Our dataset serves as a challenging new benchmark to facilitate robust and generalizable deblurring models.

Deblurring in the Wild: A Real-World Image Deblurring Dataset from Smartphone High-Speed Videos

TL;DR

This work addresses the gap in real-world image deblurring benchmarks by introducing a large-scale, smartphone-based dataset built from 240 fps slow-motion video to synthesize realistic long-exposure blur. The authors generate over 42,000 blur–sharp pairs at 1920×1080 by averaging 30 consecutive frames, with the temporally centered frame serving as ground truth, yielding an effective exposure of s. They benchmark several state-of-the-art deblurring models and observe significant performance degradation on this realistic dataset, underscoring the challenge of non-uniform, real-world blur and the need for more robust models. The dataset, captured on an iPhone 15 Pro, offers large scale, diversity, and domain relevance for training and evaluating deblurring methods aimed at consumer photography, paving the way for better generalization to practical smartphone imagery. The work highlights the real-world relevance of frame-averaging blur synthesis and motivates future research into more perceptually aligned and hardware-aware deblurring approaches.

Abstract

We introduce the largest real-world image deblurring dataset constructed from smartphone slow-motion videos. Using 240 frames captured over one second, we simulate realistic long-exposure blur by averaging frames to produce blurry images, while using the temporally centered frame as the sharp reference. Our dataset contains over 42,000 high-resolution blur-sharp image pairs, making it approximately 10 times larger than widely used datasets, with 8 times the amount of different scenes, including indoor and outdoor environments, with varying object and camera motions. We benchmark multiple state-of-the-art (SOTA) deblurring models on our dataset and observe significant performance degradation, highlighting the complexity and diversity of our benchmark. Our dataset serves as a challenging new benchmark to facilitate robust and generalizable deblurring models.

Paper Structure

This paper contains 17 sections, 4 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Image Deblurring Models
  4. Synthetic Datasets
  5. Real-world Datasets
  6. Challenges and Contributions
  7. Dataset Construction
  8. Blur Formation Model
  9. Ground Truth Definition
  10. Camera Sensor Quality
  11. Dataset Diversity and Real-World Relevance
  12. Experiments and Evaluation
  13. Benchmarks
  14. Visual Examples
  15. Conclusion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Comparison of SloMoDeblur dataset with other popular datasets.
  • Figure 2: (a) Extracting pixel data from an image frame (b) Layering temporally close frames to produce a synthetic blur
  • Figure 3: Paired images with sharp and blurred versions.