-
Download the HICO dataset from here and extract it to
./assets/data/hico -
Download our customized HICO annotations from here and extract them to
./assets/data/hico/hico_20160224_det -
Download the images of GQA from here and extract them to
./assets/data/gqa -
Download our customized GQA annotations and meta files from here and extract them to
./assets/data/gqa -
Download the pretrained vision backbones and other files from here and extract them to
./cache -
The file structure should look like
data ├── gqa │ ├── dicts.json │ ├── gqa_dic.pkl │ ├── images │ └── raw │ └── questions1.2 │ ├── test_balanced_questions.json │ ├── testdev_balanced_questions.json │ ├── train_balanced_concepts.json │ ├── train_balanced_questions.json │ ├── train_sys_reduced_concepts.json │ ├── train_sys_reduced_questions.json │ ├── val_balanced_questions.json │ └── val_sys_reduced_questions.json └── hico └── hico_20160224_det ├── images │ ├── test2015 │ └── train2015 ├── instances_test2015.json ├── instances_train2015.json ├── sys_vcl_nonrare_instances_test2015.json ├── sys_vcl_nonrare_instances_train2015.json ├── sys_vcl_rare_instances_test2015.json └── sys_vcl_rare_instances_train2015.json