search

hengyuan-hu / bottom-up-attention-vqa

An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge.
GNU General Public License v3.0
753 stars 181 forks source link

how to generates the file train_ids.pkl and val_ids.pkl #5

Closed SeekPoint closed 6 years ago

xiaoxiao26-zz commented 6 years ago

They are provided in the repo under the data folder, but if you want new ones you can perform the following steps:

  1. Download the training and validation images from http://www.visualqa.org/download.html into the data folder.
  2. Run tools/detection_features_converter.py. Make sure the old ids are deleted.

The reason the id files are provided directly in the repo is that downloading all the images in MSCOCO yourself just to grab the ids is very wasteful.