Realistic Compound-Lens Defocus Blur Synthesis

2026-07-07 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, medium

Summary

A new pipeline is proposed for synthesizing realistic defocus deblurring datasets, addressing the limitations of existing datasets that cause deep learning deblurring methods to degrade across diverse cameras and lenses. This unified pipeline integrates efficient wave-optics Point Spread Function (PSF) computation using Debye CZT propagation, depth-aware defocus rendering with occlusion handling, and blur synthesis in a radiometrically linear space with camera ISP simulation. Using this approach, the authors generated CLDefocus, a large-scale synthetic dataset featuring lens-diverse defocus image pairs. Experiments show that models trained on CLDefocus achieve improved cross-device generalization compared to those trained on current real and synthetic datasets, highlighting imperfections in real-captured data that can bias evaluation.

Key takeaway

For Machine Learning Engineers developing deblurring solutions, you should consider integrating synthetic data generation pipelines that account for diverse optical characteristics. Your models trained on datasets like CLDefocus will exhibit significantly improved cross-device generalization, reducing performance degradation when deployed across various cameras and lenses. Prioritize datasets that simulate realistic lens properties and ISP effects to overcome biases found in many real-captured datasets.

Key insights

Synthesizing diverse, realistic defocus blur datasets improves deblurring model generalization across various camera systems.

Principles

• Optical diversity is crucial for robust deblurring models.
• Real-world dataset imperfections can bias model evaluation.

Method

The pipeline combines Debye CZT propagation for PSF, depth-aware rendering with occlusion, and linear space blur synthesis with camera ISP simulation.

In practice

• Generate synthetic datasets using wave-optics PSF computation.
• Account for occlusion and camera ISP in blur synthesis.

Topics

• Defocus Deblurring
• Synthetic Data Generation
• Wave Optics
• Point Spread Function
• Deep Learning Datasets
• Cross-Device Generalization

Code references

• cidautai/DeMoE

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• How this AI Learned to Refocus Any Photo — AI refocusing allows dynamic post-capture adjustment of photographic focus and depth of field from a single image.
• Dehaze-GaussianImage: Zero-Shot Dehazing via Efficient 2D Gaussian Splatting Representation — Dehaze-GaussianImage uses 2D Gaussian Splatting for zero-shot dehazing, modeling haze as evolvable anisotropic Gaussian fields.
• Making the Discrete Continuous: Synthetic RAW Augmentations for Fine-Grained Evaluation of Person Detection Performance in Low Light — Synthetic RAW augmentation can continuously evaluate low-light person detection, overcoming real data sparsity.
• DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis — DF3DV-1K provides a large-scale, diverse dataset and benchmark for distractor-free novel view synthesis.
• Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective — Feed-forward 3D reconstruction offers efficient, generalizable scene modeling by directly mapping 2D inputs to 3D representations.
• Pixel Cube: Diffusion-based Portrait Video Relighting Through Realistic Lighting Reproduction — Diffusion models, trained on hybrid real/synthetic data from a specialized LED stage, enable photorealistic, temporally consistent portrait video relighting.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.