RaindropClarity:ADual-FocusedDatasetforDayandNightRaindropRemoval

🌈 Abstract

The paper introduces a new large-scale real-world dataset called "Raindrop Clarity" for raindrop removal research. The dataset includes 15,186 high-quality pairs/triplets of raindrop images and their corresponding clear background images, covering both daytime and nighttime scenarios, as well as raindrop-focused and background-focused images. This dataset aims to address the limitations of existing raindrop datasets, which primarily consist of daytime background-focused images with blurry raindrops. The authors demonstrate that even state-of-the-art raindrop removal methods struggle to handle the challenges presented by the new dataset, suggesting the need for further research in this area.

🙋 Q&A

[01] Dataset Characteristics

1. What are the key features of the Raindrop Clarity dataset?

• The dataset includes 15,186 high-quality pairs/triplets of raindrop images and their corresponding clear background images.
• It covers both daytime and nighttime scenarios, as well as raindrop-focused and background-focused images.
• The dataset aims to address the limitations of existing raindrop datasets, which primarily consist of daytime background-focused images with blurry raindrops.

2. How does the Raindrop Clarity dataset differ from existing raindrop datasets?

• Existing datasets mainly focus on daytime background-focused images with blurry raindrops, while Raindrop Clarity includes raindrop-focused images and nighttime raindrops.
• Raindrop Clarity provides both pairs (raindrop image and clear background) and triplets (raindrop image, blurry background, and clear background), whereas existing datasets typically only provide pairs.
• The dataset covers a wider range of raindrop sizes, shapes, and occlusion types, as well as diverse scenes and lighting conditions (both day and night).

3. What are the challenges addressed by the Raindrop Clarity dataset?

• The dataset aims to support research in eliminating raindrops and restoring clear backgrounds, regardless of whether the camera's focus is on the raindrops (resulting in a blurry background) or on the background (resulting in blurry raindrops).
• It also covers both daytime and nighttime scenarios, which present unique challenges due to differences in lighting conditions and raindrop appearances.

[02] Experimental Evaluation

1. How did the authors evaluate the performance of existing methods on the Raindrop Clarity dataset?

• The authors retrained state-of-the-art raindrop removal algorithms and restoration methods on the daytime and nighttime subsets of the Raindrop Clarity dataset.
• They used PSNR, SSIM, and LPIPS as the evaluation metrics to assess the performance of these methods.

2. What were the key findings from the experimental evaluation?

• For daytime raindrop-focused and nighttime raindrop-focused images, existing state-of-the-art methods struggled to remove raindrops and recover the background details.
• For daytime background-focused images, the methods could handle most raindrops, but struggled with long strips of raindrops.
• For nighttime background-focused images, the methods had difficulty removing raindrops and restoring clear details, as nighttime raindrops are more complex due to the influence of artificial lighting.

3. What are the implications of the experimental results?

• The experimental results suggest that there are unresolved challenges in the field of raindrop removal, particularly in handling raindrop-focused and nighttime scenarios, which have been overlooked by existing datasets and methods.
• The authors believe that further exploration of the Raindrop Clarity dataset can help the community address these more complex raindrop removal problems.

</output_format>