Skip to content

Latest commit

 

History

History
31 lines (28 loc) · 2.3 KB

download.md

File metadata and controls

31 lines (28 loc) · 2.3 KB
Raw

😇 Neutral images

For the next steps we'd need images which do not contain any known traffic signs. We leverage OpenimagesV5 and build a neutral image set by querying for Building and filtering out images containing traffic signs, referred below as Building_without_signs.list. Please refer to instructions on Figure8.. Please download the following file(s) and place them under synthetic_signs/external/lists:

😐 Neutral images list gen

python scripts/download-openimages-v5.py --help
Download Class specific images from OpenImagesV5

optional arguments:
  -h, --help           show this help message and exit
  --mode MODE          Dataset category - train, validation or test
  --classes CLASSES    Names of object classes to be downloaded
  --nthreads NTHREADS  Number of threads to use
  --dest DEST          Destination directory
  --csvs CSVS          CSV file(s) directory
  --limit LIMIT        Cap downloaded files to limit

Sample command

python scripts/download-openimages-v5.py --classes 'Building,Traffic_sign,Traffic_light' --mode train --dest <path_to_openimages_v5> --csvs synthetic_signs/external/lists

Build class list and filter image set & Filtering for outdoor images with no labeled signs.

find path_to_openimages_v5/Building -type f -name '*.jpg' > synthetic_signs/external/lists/Building.list
find path_to_openimages_v5/Traffic_sign -type f -name '*.jpg > synthetic_signs/external/lists/Traffic_sign.list
find path_to_openimages_v5/Traffic_light -type f -name '*.jpg > synthetic_signs/external/lists/Traffic_light.list