The RealDefocus Benchmark for Defocus Deblurring

2026-07-23

AI Summary

Current research on single-image defocus deblurring is hindered by the lack of real-world, high-resolution, and precisely aligned paired datasets, as well as a unified evaluation protocol, making fair comparison and reproducibility difficult. This work introduces RealDefocus, a new benchmark that, for the first time, extracts high-quality defocused–sharp image pairs from authentic bokeh imaging data. It establishes standardized data splits and a consistent training/validation/testing pipeline, and further proposes a cross-dataset evaluation protocol to assess both reconstruction fidelity and generalization capability. RealDefocus provides a reproducible and reliable platform for the defocus deblurring community, significantly advancing fair algorithmic comparison and performance improvement.

Abstract

Single-Image Defocus Deblurring (SIDD) aims to recover an all-in-focus image from a single defocused observation, but rigorous and reproducible evaluation remains challenging due to the scarcity of realistic, high-resolution datasets with well-aligned defocused/sharp pairs and standardized protocols. We build on RealDefocus, a benchmark derived from the real-world RealBokeh dataset originally proposed for Bokeh Rendering. RealDefocus provides paired defocused inputs and sharp ground truth images, predefined training/validation/test splits, and a unified evaluation framework for comparing image restoration and neural rendering approaches. We further outline a benchmarking protocol with cross-dataset validation to assess reconstruction quality and generalization. The project page is publicly available at: www.github.com/TimSeizinger/RealDefocus-Benchmark.

Problem

• Defocus Deblurring
• Single-Image
• Benchmark
• Realistic Dataset
• Evaluation Protocol

Innovation

• Defocus Deblurring
• RealDefocus Benchmark
• Single-Image Defocus Deblurring
• Cross-dataset Validation
• Unified Evaluation Framework