The New Trends in Image Restoration and Enhancement (NTIRE) 2023 workshop and associated challenges are international competitions aiming to promote the development of computer vision and image processing. They are held in conjunction with the Institute of Electrical and Electronics Engineers (IEEE) / Computer Vision Foundation (CVF) Computer Vision and Pattern Recognition Conference (CVPR). The competition focuses on image denoising, super-resolution, deblurring, inpainting, colorization, etc. In addition, it aims to showcase new techniques that can improve image quality and enhance details, texture, and visual appearance.

We are proud to announce that Samsung R&D Institute China - Beijing (SRC-B) has recently been crowned as champions in the following tracks at the NTIRE 2023:

• Bokeh Effect Transformation (Champion)

• HR Depth from Images of Specular and Transparent Surfaces – Stereo (Champion)

• Stereo Image Super-Resolution – Track 2 Perceptual & Bicubic (Champion)

• Image Denoising (Runner-up)

The “Bokeh Effect Transformation” track challenged participants to transform a given bokeh image into another bokeh effect image, simulating the blurred background created by camera lenses when taking portraits. SRC-B utilized a sophisticated deep learning architecture equipped with a novel attention mechanism to generate stunning bokeh effect images that outperformed other competing methods.

The “HR Depth from Images of Specular and Transparent Surfaces” track challenged participants to estimate high-quality depth maps from images of specular and transparent surfaces. SRC-B leveraged a novel multi-scale and multi-task deep learning model to effectively address this problem, delivering exceptional performance compared to all other participating teams.

The “Stereo Image Super-Resolution” track focused on generating high-resolution stereo images from low-resolution inputs. SRC-B employed the Swin Transformer and Fast Fourier Convolution mechanism to achieve state-of-the-art performances on this challenging task.

The “Image Denoising” track required participants to remove noise from a set of corrupted images with various noise levels. SRC-B developed a unique approach that combined feature ensemble algorithms with a deep neural network to produce impressive results with superior perceptual quality.

SRC-B has been performing well in computer vision tasks, aiming at image restoration, enhancement, and manipulation, and winning the challenges at NTIRE workshops for three consecutive years from 2021 to this year. During the competition, SRC-B faced fierce competition from top players across the globe. After months of hard work and dedication to perfecting their techniques, they successfully overcame various challenges in the competition through innovative methods and techniques and ultimately won the championship.

SRC-B’s team members