Ian824/High-Resolution-Rainy-Image

The High-resolution Rainy Image (HRI) dataset is a synthetic dataset created through a learning-from-rendering approach, detailed in the paper Learning from Rendering: Realistic and Controllable Extreme Rainy Image Synthesis for Autonomous Driving Simulation. Designed for autonomous driving simulation, HRI provides realistic and controllable extreme rainy images to enhance visual perception models. It comprises 3,200 paired rainy-clean images, along with corresponding depth and rain layer mask images, captured across three diverse scenes (lane, citystreet, and japanesestreet) at a high resolution of 2048x1024. This dataset is particularly valuable for tasks such as semantic segmentation, instance segmentation, depth estimation, and object detection in challenging weather conditions, as demonstrated in the related work with the CARLARain simulator.

High-resolution Rainy Image (HRI) Dataset

This is the dataset in the paper "Learning from Rendering: Realistic and Controllable Extreme Rainy Image Synthesis for Autonomous Driving Simulation".

• Project Page: https://kb824999404.github.io/HRIG/
• Paper: Learning from Rendering: Realistic and Controllable Extreme Rainy Image Synthesis for Autonomous Driving Simulation
• Code (Learning-from-Rendering): https://github.com/kb824999404/HRIG
• Code (CARLARain): https://github.com/kb824999404/CARLARain

HRI Dataset

The High-resolution Rainy Image (HRI) dataset in the rendering stage.

scene	dataset type	resolution	viewpoints	moments	intensities	image pairs
lane	training set	2048×1024	3	100	4	1200
	test set		1			400
citystreet	training set	2048×1024	5	25	4	500
	test set		1			100
japanesestreet	training set	2048×1024	8	25	4	800
	test set		2			200

• clean: background RGB images and depth images of all scenes.
• rainy: rain layer images, RGB rainy images and depth rainy images of all scenes.
• trainset.json: the sample lists of the training set.
• testset.json: the sample lists of the test set.
• For each sample in the training set and the test set:
  • scene: the scene name
  • sequence: the viewpoint name
  • intensity: the rain intensity
  • wind: the wind direction( all zero for the HRI dataset)
  • background: the path of the background RGB image
  • depth: the path of the background depth image
  • rain_layer: the path of the rain layer image
  • rainy_depth: the path of the rainy depth image
  • rainy_image: the path of the rainy RGB image

BlenderFiles

The Blender files for rendering RGB and depth images of all viewpoints are included in the directory of each scene.

CARLARain-Data

• ExtremeRain: Based on CARLARain, we construct an extreme rainy street scene image dataset, ExtremeRain. This dataset contains 8 different street scenes and 3 illumination conditions: daytime, sunset, night. The rainy scenes feature a rain intensity ranging from 5 mm/h - 100 mm/h, covering extreme rainfalls under complex illumination conditions. The dataset contains comprehensive label information to meet the requirements of multi-task visual perception models, including semantic segmentation, instance segmentation, depth estimation, and object detection. We split the dataset into train set and test set according to different scenes.

Rain streak database

The Rain streak database from the paper Rain Rendering for Evaluating and Improving Robustness to Bad Weather.

• rain-streak-database.zip: all rain streak images from the rain streak database
• rain-mask-wind.tar.gz: precreated rain mask images with different intensity and direction

Citation

When using these datasets, please cite our paper:

@article{zhou2025high,
  title={Learning from Rendering: Realistic and Controllable Extreme Rainy Image Synthesis for Autonomous Driving Simulation},
  author={Kaibin Zhou, Kaifeng Huang, Hao Deng, Zelin Tao, Ziniu Liu, Lin Zhang, Shengjie Zhao},
  journal={arXiv preprint arXiv:2502.16421},
  year={2025}
}