dataset_prepare.md

HazeWorld

HazeWorld is a large-scale synthetic outdoor video dehazing dataset, which is built upon Cityscapes, DDAD, UA-DETRAC, VisDrone, DAVIS, and REDS. Please refer to these official dataset websites for rights of use.

We use RCVD to estimate the temporally consistent video depths, which are used to synthesize the hazy videos. The fog synthesis pipeline is built on SeeingThroughFog.

Prepare datasets

It is recommended to symlink the dataset root to $MAP-NET/data. If your folder structure is different, you may need to change the corresponding paths in config files.

MAP-Net
├── ...
├── data
│   ├── Cityscapes
│   │   ├── leftImg8bit_sequence_trainvaltest
│   │   │   ├── leftImg8bit_sequence
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── test
│   ├── HazeWorld
│   │   ├── mapping_hazeworld_cityscapes.txt
│   │   ├── gt
│   │   │   ├── Cityscapes
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── test
│   │   │   │   ├── mapping_info_GT_train.txt
│   │   │   │   ├── mapping_info_GT_test.txt
│   │   │   ├── DDAD
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── ...
│   │   │   ├── UA-DETRAC
│   │   │   │   ├── train
│   │   │   │   ├── test
│   │   │   │   ├── ...
│   │   │   ├── VisDrone
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── test-dev
│   │   │   │   ├── ...
│   │   │   ├── DAVIS
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── test-dev
│   │   │   │   ├── test-challenge
│   │   │   │   ├── ...
│   │   │   ├── REDS
│   │   │   │   ├── train
│   │   │   │   ├── val
│   │   │   │   ├── ...
│   │   ├── hazy
│   │   │   ├── ...
│   │   ├── transmission
│   │   │   ├── ...
(symlink)
│   │   ├── train
│   │   │   ├── meta_info_tree_GT_train.json
│   │   │   ├── meta_info_GT_train.txt
│   │   │   ├── meta_info_GT_...
│   │   │   ├── gt (symlink)
│   │   │   │   ├── Cityscapes
│   │   │   │   ├── DDAD
│   │   │   │   ├── UA-DETRAC
│   │   │   │   ├── VisDrone
│   │   │   │   ├── DAVIS
│   │   │   │   ├── REDS
│   │   │   ├── hazy (symlink)
│   │   │   │   ├── ...
│   │   ├── test
│   │   │   ├── meta_info_tree_GT_test.json
│   │   │   ├── meta_info_GT_test.txt
│   │   │   ├── meta_info_GT_...
│   │   │   ├── gt (symlink)
│   │   │   │   ├── ...
│   │   │   ├── hazy (symlink)
│   │   │   │   ├── ...

Step 1. Download the data from the links at the bottom. Since many hazy videos may correspond to one ground-truth video, we adopt the file structure above to save storage.

Step 2. Download the meta files and put them into the corresponding locations (see above). The meta files are provided in the HazeWorld download link.

Step 3. Symlink the train and test split using the script and the following command:

python tools/data/dehazing/hazeworld/create_symlink_hazeworld.py

Cityscapes

The data could be found here after registration. Download leftImg8bit_sequence_trainvaltest.zip (324GB). The used videos can be found in mapping_hazeworld_cityscapes.txt.

python tools/data/dehazing/hazeworld/preprocess_hazeworld_cityscapes.py \
--meta-file data/HazeWorld/mapping_hazeworld_cityscapes.txt \
--input-dir data/Cityscapes --work-dir data/HazeWorld

Others

For others, we provide the processed data (~100GB). You can also refer to their official websites for the original data.

Notes

• We do some data processing on the original data, so the numbers and videos may not correspond to the original ones. Here are some examples, and more details can be found here.

  We sample the frames to keep each video clip of similar length (mostly no more than 100 images per video), using different sampling strategies for each dataset.

  We use cv2 to resize (short border to 720 pixels if the original is larger than 720, keeping aspect ratios, and using the default interpolation method) and save (jpg, default quality, lossy compression, to save storage) images.

• Also, we manually check the data and remove some improper videos (e.g., indoor or nighttime scenes).