• Synthetic rainy datasets for single image deraining are collected here.

Single Image Datasets

Synthetic Datasets

• Rain12_CVPR2016 in LP

• Rain100L_CVPR2017 in JORDER

• Rain100H_CVPR2017 new version in JORDER

  • Note that Rain100H provided by Yang et al. is updated in their website. Original version contains 1800 clean/rainy training image pairs and 100 clean/rainy testing image pairs. The new version contains 1800 clean/rainy training image pairs and 200 clean/rainy testing image pairs. All the methods assess their performances on original version instead of the new one. As a result, the metrics will be different when you retrain the models using the new one.

• Rain100H_CVPR2017 old version:

  • train set
    • Link: https://pan.baidu.com/s/1uiTz_ME_9Lq5k0yA6c7ZEw
    • Extraction code: x3gy
  • test set
    • Link: https://pan.baidu.com/s/1GLLpNJ8P4mkwfX7meMphTg
    • Extraction code: axpq

• Rain800_Arxiv2017 in ID_CGAN

• Rain1400_CVPR2017 in DDN

• Rain1200_CVPR2018 in DID_MDN

• SPANet Dataset__CVPR2019 in SPANet

• Heavy Rainy Dataset_CVPR2019 in Heavy Rain Image Restoration

Real-World Datasets

• Practical in JORDER

• Practical in ID_CGAN

Synthetic Methods

• Photoshop