NTIRE 2024 Challenge on Light Field Image Super-Resolution: Methods and Results

In this report, we summarize the 2nd NTIRE challenge on light field (LF) image super-resolution (SR) with a focus on new methods and results. This challenge aims at superresolving LF images under the standard bicubic downsampling degradation with a magnification factor of ×4. Compared with single image SR, the major challenge of LF image SR lies in how to exploit complementary angular information from plenty of views with varying disparities. This year of challenge has two tracks, including one track on fidelity (i.e., restoration accuracy in terms of PSNR) only, and the other track on fidelity with an extra constraint on model size and computational cost. In total, 125 participants were successfully registered for this challenge, and 9 teams have successfully submitted results with PSNR scores higher than the baseline methods. We report the solutions proposed by the participants, and summarize their common trends and useful tricks. We hope this challenge can stimulate future research and inspire new ideas in LF image SR.

1. Introduction

Light field (LF) cameras are capable of capturing the intensity and directions of light rays, allowing for the recording of 3D geometry in a practical and effective way. Through encoding 3D scene cues into 4D LF images (2D for spatial dimension and 2D for angular dimension), LF cameras can facilitate numerous appealing applications, ineluding post-capture refocusing [1], [2], depth sensing [3–5], virtual reality [6], [7], and view rendering [8]–[11].