Annonce

Les commentaires sont clos.

Restore Any Image Model (RAIM) in the Wild: An NTIRE Challenge in Conjunction with CVPR 2024

8 Février 2024

Catégorie : Compétitions et challenges

The 9th edition of NTIRE: New Trends in Image Restoration and Enhancement workshop will be held in June 2024 in conjunction with CVPR 2024.

1. Motivation

Image restoration, aiming at recovering high-quality images from their low-quality counterparts, is one of the most popular low level vision tasks in the research community. However, there has been a large gap between the Academic research and the Industrial application for a long time. For example, the image signal processing (ISP) systems on digital cameras always face mixed and complex degradations, yet most methods in academic research are designed and evaluated based on simulated and limited degradations. How to design and train a model which can generalize to practical applications is a challenging yet highly valuable problem.

The deep learning techniques have significantly advanced the performance of image restoration. Recently, the large scale pretrained generative diffusion models have provided powerful priors to further improve the quality of image restoration outputs. To provide a platform for researchers to investigate how to bridge the gap between academic research and industrial application, the Y-Lab of The OPPO Research Institute and the Visual Computing Lab of The Hong Kong Polytechnic University co-host this challenge of Restore Any Image Model (RAIM) in the Wild. In this challenge, we will provide comprehensive data collected in real-world digital photography for researchers to test their models, as well as high-quality feedback from experienced practitioners in industry.

2. Objectives

This challenge aims to provide a platform for the industrial and academic participants to test and evaluate their algorithms and models on real-world imaging scenarios, bridging the gap between academic research and practical photography. The objectives of this RAIM challenge are:

Construct a benchmark for image restoration in the wild, including real-world images with/without reference ground-truth in various scenarios and objective/subjective evaluation methods;
Promote the research and development of RAIMs with strong generalization performance to images in the wild.

3. Awards

The following awards of this challenge are provided:

One first-class award (i.e., the champion) with case prize US$1000;
Two second-class awards with case prize US$500 each;
Three third-class awards with case prize US$200 each.

4. Phases

4.1 Phase 1: Model Design and Tuning

In this phase, participants can analyze the given data and tune their models accordingly. We will provide:

100 pairs of paired data (i.e., input with R-GT), which can be used to tune the models based on the quantitative measures.
100 images without R-GT, which can be used to tune the model according to visual perception.

4.2 Phase 2: Online Feedback

In this phase, participants can upload their results and get official feedback. We will provide:

The input low-quality images of another 100 pairs of paired data. Only the low-quality input images are provided, and the participants can upload the restoration results to the server and get the quantitative scores online.
Users can also upload their results of the images without R-GT provided in Phase 1 to seek feedback. The organizers will provide feedback to a couple of teams that get the highest quantitative scores of the images with R-GT.

4.3 Phase 3: Final Evaluation

In this phase, we will provide:

Another 50 images without R-GT for subjective evaluation.

In this phase, we select the top ten teams according to the quantitative score of the 100 images with R-GT in Phase 2, and then arrange a comprehensive user study on their results of the above 50 images without R-GT. The final ranks of the ten teams will be decided based on both the quantitative scores and the subjective user study (the weight will be given later).