Raindrops on camera lenses or windshields significantly reduce image visibility in human life, posing challenges for applications like surveillance and autonomous driving. Effective raindrop removal is crucial to ensuring reliable performance in these systems. However, most deraining datasets overlooked the complex environment in real life. In particular, there are few raindrop-focused datasets, and most existing datasets focus on capturing background scenes while the camera is focused on the background. Meanwhile, these datasets primarily target daytime scenarios, with limited attention to nighttime conditions.

To promote the development of deraining, we introduce the first large-scale real-world raindrop-clarity dataset, which includes raindrop-focused images and nighttime scenes. Based on this dataset, we have the first NTIRE Challenge on Day and Night Raindrop Removal for Dual-Focused Images, jointly with the NTIRE workshop.

The goal of this competition is to establish a new and applicable benchmark for the Day and Night Raindrop Removal for Dual-Focused Images. We are looking forward to the collaborative efforts of our participants, aiming to elevate the quality of training images. The final ranking will be made based on PSNR, SSIM, and LPIPS.

The top-ranked participants will be awarded and invited to follow the CVPR submission guide for workshops to describe their solution and to submit to the associated NTIRE workshop at CVPR 2025.

• 2025.01.28 Release of train data (input and output)
• 2025.02.01 Release of validation data (inputs only)
• 2025.02.01 Validation server online
• 2025.03.15 Final test data release (inputs only)
• 2025.03.21 Test output results submission deadline
• 2025.03.22 Fact sheets and code/executable submission deadline
• 2025.03.24 Preliminary test results release to the participants
• 2025.04.05 Paper submission deadline for entries from the challenge
• 2025.06.??? NTIRE workshop and challenges, results, and award ceremony (CVPR 2025, Nashville, USA)

The dataset named RainDrop Clarity is contributed by Yeying Jin et al (NUS). The dataset includes both daytime and nighttime scenes for training and testing. (i) The daytime raindrop dataset contains a total of 5,442 paired or triplet images, with 4,713 pairs used for training and the remaining 729 pairs for testing. (ii) The nighttime raindrop dataset consists of 9,744 paired or triplet images, where 8,655 pairs are allocated to the training set, and the remaining 1,089 pairs are reserved for the validation and test set.

In this competition, the used dataset is composed of: