We present a video rain dataset that contains both sythetic rain (by Adobe After Effect) and real rain video clips for algorithm training and testing.
[dataset download]
[paper]
[code]
The data category usage (i.e., training/testing dataset), the rain types (i.e., whether the rain is synthetic/real rain), the camera motion for each data entry (i.e., slow/fast moving cameras), as well as the labeling details for each data is shown in the table below.
We show the thumbnails for each data entry below.
Please cite both Adobe After Effect [2], and our paper [1] when you use this dataset, or compare with our results.
Organization structure details of the dataset:
-
Under the folder “SPAC-CNN_code” folder are the executable code for rain removal.
- Under the “Dataset_Testing_Synthetic” and “Dataset_Training_Synthetic” folders are the synthetic rain dataset for model training, and algorithm testing.
- Files with names xxx_GT are the scene ground truth.
- Files with names xxx_Rain are the synthesized rainy scenes with Adobe Affter Effect.
- Files with names xxx_spacCNN are the derain output from SPAC-CNN (only for the testing dataset).
Please cite both Adobe After Effect [2], and our paper [1] when you use this dataset, or compare with our results.
- Under the “Dataset_Testing_realRain” folder are the real rain datset for evaluation of derain performance
- Files with names xxx_Rain are the rainy frames.
- Files with names xxx_spacCNN are the derain output from SPAC-CNN.
Please cite our paper [1] when you use this dataset, or compare with our results.
-
Under the “videoDemo” folder are the rain removal video clips used for evaluations in the paper:
The 8 video files starting with “synthetic_” are the derain results for the synthetic rain dataset used for quantitative evaluations in our paper.
- The 4 videos with names “Synthetic_slow_axxx.mp4” belongs to group a.
- The 4 videos with names “Synthetic_fast_bxxx.mp4” belongs to group b.
The 7 video files starting with “realrain_” are derain results for real rain.
- Under the “SupperPixel_Matching_Evaluations” foder are the experiment results of content matching using SP units vs. rectangular patches for the Middleburry Stereo Dataset. Matching results for the SP units are with names “xxx_sp.png”, and the results for the rectangular patches are with names “xxx_blk.png”.
[1] J. Chen, C.-H. Tan, J. Hou, L.-P. Chau, and H. Li, “Robust video content alignment and compensation for rain removal in a CNN framework,” IEEE Conference on Computer Vision and Pattern Recognition, 2018. [pdf] [code]
[2] Adobe After Effects Software, available at www.adobe.com/AfterEffects.