Data Preparation

Download Images

Download COCO image dataset (for VQA-X)

wget http://images.cocodataset.org/zips/train2014.zip
wget http://images.cocodataset.org/zips/val2014.zip

Download Annotations

COCO Caption for evaluation

git clone https://github.com/tylin/coco-caption.git
mv coco-caption cococaption
cd cococaption
./get_stanford_models.sh
pip install bert_score==0.3.7

Download VQA-X annotations

gdown https://drive.google.com/drive/folders/16sJjeEQE2o23G-GGUi870ubXzJjdRDua --folder
mv VQA-X nle_anno/VQA-X

Download processed annotations

gdown https://drive.google.com/drive/folders/1b8kUPbgtEduiz8A_VbUg0W_vca7PyXsZ --folder
mv "cococaption annot"/* cococaption/annotations/ 

Download Captioning dataset

nocaps <validation set, test set>

wget https://nocaps.s3.amazonaws.com/nocaps_val_4500_captions.json

Localized Narratives <train image set(10), validation image set , test image set, train caption set, validation caption set, test caption set > image data from open-images-dataset
https://github.com/cvdfoundation/open-images-dataset.git

wget https://storage.googleapis.com/localized-narratives/annotations/open_images_train_v6_captions.jsonl
wget https://storage.googleapis.com/localized-narratives/annotations/open_images_validation_captions.jsonl
wget https://storage.googleapis.com/localized-narratives/annotations/open_images_test_captions.jsonl

pip install jsonlines

COCO-caption
https://github.com/tylin/coco-caption.git