vg_preprocessing

Visual Genome dataset preprocessing

1. Data

Download region_descriptions.json.zip and extract it.

2. Usage

Just for test, 'imgs' folder contains only first 200 images from the original dataset(images.zip)

1. '1_img_annotating.py' creates 'imgs_anno' folder to annotate the original images using the information of 'region_descriptions.json'.

2. '2_json_pruning.py' creates 'region_descriptions_pruned.json' file, which eliminates many unnecessary duplicated annotations.