8 February 2024
Computer Vision and Pattern Recognition
New UAV-Rain1k benchmark dataset for raindrop removal from drone images
800 train and 220 test images with diverse rain densities and aerial angles
Evaluated CNN and transformer deraining algorithms, transformers performed best
Models still struggle with some artifacts and local detail preservation
Removing Raindrops from Drone Images
This paper introduces a new benchmark dataset called UAV-Rain1k for removing raindrops from drone aerial images. It contains 800 training and 220 test images with simulated raindrops composited onto real drone footage backgrounds. The dataset has diverse rain densities and aerial shooting angles to better represent real-world challenges. The authors evaluate several state-of-the-art deraining algorithms on UAV-Rain1k, showing transformers can outperform CNNs but still struggle with some artifacts and detail preservation. They conclude by calling for more research into raindrop removal tailored for drone imagery.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper