multi-task-defocus-deblurring-dual-pixel-nimat
Improves single-image defocus deblurring by learning from dual-pixel images in a multi-task framework
Reference github repository for the paper "Improving Single-Image Defocus Deblurring: How Dual-Pixel Images Help Through Multi-Task Learning". We propose a single-image deblurring network that incorporates the two sub-aperture views into a multitask framework. Specifically, we show that jointly learning to predict the two DP views from a single blurry input image improves the network’s ability to learn to deblur the image. Our experiments show this multi-task strategy achieves +1dB PSNR improvement over state-of-the-art defocus deblurring methods. In addition, our multi-task framework allows accurate DP-view synthesis (e.g., ~ 39dB PSNR) from the single input image. These high-quality DP views can be used for other DP-based applications, such as reflection removal. As part of this effort, we have captured a new dataset of 7,059 high-quality images to support our training for the DP-view synthesis task.
Related projects:
| Repository | Description | Stars |
|---|---|---|
| abdullah-abuolaim/defocus-deblurring-dual-pixel | Developing a deep learning model to correct blurry images caused by camera shake or out-of-focus | 185 |
| abdullah-abuolaim/recurrent-defocus-deblurring-synth-dual-pixel | This project provides tools and models to generate realistic data for camera systems with defocus blur, aiming to improve image deblurring techniques. | 49 |
 hyeongseokson1/kpac | An implementation of a deep learning model for deblurring images affected by defocus. | 58 |
 rozumden/defmo | A deep learning framework for deblurring and recovering the shape of fast-moving objects from blurred images | 170 |
 radimspetlik/si-ddpm-fmo | A Python-based framework for training and evaluating deep learning models for single-image deblurring, shape, and trajectory recovery of fast-moving objects. | 5 |
 minyuanye/siun | This project develops a deep learning-based image deblurring algorithm using iterative upsampling network architecture | 162 |
 wdzhao123/apl | Develops a deep learning-based method to detect and remove defocus blur from images | 16 |
 deepmed-lab-ecnu/deeprft-aaai2023 | A deep learning-based image deblurring system that explores the impact of frequency selection on restoration quality | 18 |
 ysnan/vem-nbd | Provides pre-trained models and benchmark results for noise-blind image deblurring, allowing developers to test and compare different approaches. | 14 |
 codeslake/ifan | Implementation of an algorithm for single image deblurring in images with defocus blur | 227 |
 chosj95/mimo-unet | Develops a deep learning model for single image deblurring with improved performance and computational efficiency | 373 |
 jihyongoh/demfi | Develops algorithms to restore sharp images from blurry ones and interpolate missing frames in video sequences with improved accuracy | 81 |
 fangzhenxuan/ufpdeblur | An image deblurring algorithm that leverages flow-based motion prior and kernel estimation for blind image restoration. | 28 |
 csjcai/dbcpenet | Deblurring technique developed using machine learning and signal processing algorithms to restore images from blurry conditions. | 20 |
 cszn/ircnn | This project trains deep CNN denoisers to improve image restoration tasks such as deblurring and demosaicking through model-based optimization methods. | 600 |