Skip to content

Latest commit

 

History

History
24 lines (22 loc) · 1.89 KB

data.md

File metadata and controls

24 lines (22 loc) · 1.89 KB

Data pre-procseeing

  • The 'pre-processing' indicates to compute and to save image features before training and testing as written in our paper.
  • The pre-processed data we provide in this link are processed following the VIBE repository's instruction.
  • Download the data from sources following the instruction, and run our python scripts as below
python lib/data_utils/{dataset_name}_utils.py --dir ./data/{dataset_name}
  • You may need to change details (ex. scale), so check comments in {dataset_name}_utils.py files.
  • To get occluusion-augmented train data following our paper (inspired by this paper),
    • First, download VOC2012 data and locate under ${ROOT}/data/.
    wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
    tar -xf VOCtrainval_11-May-2012.tar
    • Second, uncomment load_occluder function and pass the occluders variable to extract_features function in {dataset_name}_utils.py.
    • Last, set occ argument in get_single_image_crop function in _feature_extractor.py.
  • We did not pre-processed InstarVariety dataset. We used the data pre-processed by VIBE.

If you have a problem with 'download limit' when trying to download datasets from google drive links, please try this trick.

  • Go the shared folder, which contains files you want to copy to your drive
  • Select all the files you want to copy
  • In the upper right corner click on three vertical dots and select “make a copy”
  • Then, the file is copied to your personal google drive account. You can download it from your personal account.