This repository is for preprocessing OpenImageSet Detection data to Pascal VOC format.
You can make VOC formatted detection dataset on OpenImage by specifying the class labels! :)
You can download annotation file and class names (.csv) from this url
I edited class name files by adding id and name like this
id,name
/m/011k07,Tortoise
/m/011q46kg,Container
/m/012074,Magpie
/m/0120dh,Sea turtle
/m/01226z,Football
/m/012n7d,Ambulance
/m/012w5l,Ladder
pandas
boto3
tqdm
imagesize
pascal_voc_writer
python make_set.py
This code makes set.txt file containing which image to download
python downloader.py set.txt --download_folder=[your path] --num_processes=5
This code download the images
If the image is not downloaded fully, then try this:
python check_missing.py
python downloader.py set2.txt --download_folder=[your path] --num_processes=5
For annotation,
python annot.py
Please change the ROOT path variable. This code makes the annotation file
if the annotation is not fully done, you can resume it by
python annot.py --resume