Skip to content

Latest commit

 

History

History
41 lines (38 loc) · 2.04 KB

prepare_data.md

File metadata and controls

41 lines (38 loc) · 2.04 KB

Data Preparation

We support 4 popular settings in DAOD as listed below:

Source Target Test
normal to foggy (C2F) cityscapes (train) cityscapes-foggy (train) cityscapes-foggy (val)
small to large (C2B) cityscapes (train) BDD100K (train) BDD100K (val)
across cameras (K2C) KITTI (train-car) cityscapes (train-car) cityscapes (val-car)
synthetic to real (S2C) Sim10K (train-car) cityscapes (train-car) cityscapes (val-car)

All datasets are aranged in the format of PASCAL VOC as follows:

# cityscapes          
- VOC2007_city 
    - ImageSets  
    - JPEGImages  
    - Annotations  

Here are datasets used in this paper.

SPLITS = [
        ("VOC2007_citytrain", 'data/VOC2007_citytrain', "train", 8),
        ("VOC2007_foggytrain", 'data/VOC2007_foggytrain', "train", 8),
        ("VOC2007_foggyval", 'data/VOC2007_foggyval', "val", 8),
        ("VOC2007_citytrain1", 'data/VOC2007_citytrain1', "train", 1),
        ("VOC2007_cityval1", 'data/VOC2007_cityval1', "val", 1),
        ("VOC2007_bddtrain", 'data/VOC2007_bddtrain', "train", 8),
        ("VOC2007_bddval", 'data/VOC2007_bddval', "val", 8),
        ("VOC2007_kitti1", 'data/kitti', "train", 1),
        ("VOC2007_sim1", 'data/sim', "train", 1),
    ]

CitysScape and FoggyCityscape

  1. register and download from CitysScape to data/
  2. transform segmentation annotations to detection formats
    • For multi classes, python tools/trans_seg_to_det_multi.py
    • For single class, python tools/trans_seg_to_det_multi.py
  3. check the annotations and make txt, python tools/make_VOC_txt.py

KITTI, Sim10k, BDD100k

Download from KITTI, Sim10k, BDD100k to data/.