< Back to Yunfei's home page

PBRR:  Physically Based Raindrop Rendering

Yunfei Liu1, Zhixiang Hao1, Shadi You2, Yu Li3, Feng Lu1

1. Beihang University

2. University of Amsterdam

3. Tencent

PBRR is a large scale, public dataset for raindrop removal of photo-realistic adherent raindrop images based on Cityscapes dataset. It contains two sub-dataset. In particular, the PBRR-Spheric dataset contains14875 data-samples using spherical crown mode; and the PBRR-Bézier dataset includes 14875 data-samples using Orthogonal Bezier Curve model. Each data sample contains images with 50-70 raindrops with various but realistic appearances, a ground truth raindrop mask, and a ground truth raindrop-less image.

Data examples



  1. Porav, H., Bruls, T., Newman, P.: I can see clearly now: Image restoration via de-raining. arXiv preprint arXiv:1901.00893(2019)
  2. Qian, R., Tan, R.T., Yang, W., Su, J., Liu, J.: Attentive generative adversarial network for raindrop removal from a single image. In: IEEE Conference on Computer Vision and Pattern Recognition(2018)



The dataset can be accessed from here.