If you use the RaidaR dataset in your work, please cite our publication:
Jiongchao Jin, Arezou Fatemi, Wallace Lira, Fenggen Yu, Biao Leng, Rui Ma, Ali Mahdavi-Amiri, and Hao Zhang. "RaidaR: A Rich Annotated Image Dataset of Rainy Street Scenes". arXiv e-prints, abs/2104.04606, April 2021. [bibtex]
Our dataset is divided into 2 folders containing the dataset splits: train and test. Each split contains 6 folders, where sseg denotes semantic segmentation and iseg denotes instance segmentation. The directory organization is shown below:
RaidaR
|---train/test
|---sunny_sseg
|---sunny_iseg
|---sunny_images
|---rainy_sseg
|---rainy_iseg
|---rainy_images
Name |
ID |
Has Instance Segmentation |
Color |
|---|---|---|---|
| Unlabeled | 0 | False | ( 0, 0, 0) |
| Static | 4 | False | ( 0, 0, 0) |
| Dynamic | 5 | False | (111, 74, 0) |
| Ground | 6 | False | ( 81, 0, 81) |
| Road | 7 | False | (128, 64,128) |
| Sidewalk | 8 | False | (244, 35,232) |
| Parking | 9 | False | (250,170,160) |
| Rail track | 10 | False | (230,150,140) |
| Building | 11 | False | ( 70, 70, 70) |
| Wall | 12 | False | (102,102,156) |
| Fence | 13 | False | (190,153,153) |
| Guard rail | 14 | False | (180,165,180) |
| Bridge | 15 | False | (150,100,100) |
| Tunnel | 16 | False | (150,120, 90) |
| Pole | 17 | False | (153,153,153) |
| Pole group | 18 | False | (153,153,153) |
| Traffic light | 19 | False | (250,170, 30) |
| Traffic sign | 20 | False | (220,220, 0) |
| Vegetation | 21 | False | (107,142, 35) |
| Terrain | 22 | False | (152,251,152) |
| Sky | 23 | False | ( 70,130,180) |
| Person | 24 | True | (220, 20, 60) |
| Rider | 25 | True | (255, 0, 0) |
| Car | 26 | True | ( 0, 0,142) |
| Truck | 27 | True | ( 0, 0, 70) |
| Bus | 28 | True | ( 0, 60,100) |
| Caravan | 29 | True | ( 0, 0, 90) |
| Trailer | 30 | True | ( 0, 0,110) |
| Train | 31 | True | ( 0, 80,100) |
| Motorcycle | 32 | True | ( 0, 0,230) |
| Bicycle | 33 | True | (119, 11, 32) |
All images and masks are in 1080p resolution. The masks and raw images share the same name, differing only on the folder they are located on.
Our dataset follows the same labeling policy ast the Cityscapes Dataset. Labeled foreground objects must never have holes，i.e. if there is some background visible ‘through’ some foreground object, it is considered to be part of the foreground.
The data loading is similar to the Cityscapes Dataset.