Skip to content

Latest commit

 

History

History
60 lines (51 loc) · 2.33 KB

PREPARE_NAQ_DATASETS.md

File metadata and controls

60 lines (51 loc) · 2.33 KB

Preparing NaQ datasets for augmentation

We provide two modes of creating the NaQ dataset.

  1. [RECOMMENDED] Download pre-generated dataset used for our paper experiments using the ego4d api.
  2. Create NaQ datsets from scratch.

See instructions for both options below.

Option 1: Download pre-generated NaQ datasets

  • We share pre-generated datasets through the Ego4D api. 
    ego4d --output_directory="<OUTPUT DIRECTORY>" --datasets naq_datasets --aws_profile_name <AWS-PROFILE-WITH-EGO4D-ACCESS> --version v2
  • Copy the data $NAQ_ROOT and extract. 
    cp `<OUTPUT DIRECTORY>/v2/baselines/naq_datasets.zip` $NAQ_ROOT/
unzip naq_datasets.zip

Option 2: Creating NaQ datasets from scratch

  • Download the Ego4D narrations in the EgoClip format provided here to $NAQ_ROOT/data/egoclip.csv.
  • Convert EgoClip narrations to NaQ dataset. 
    cd $NAQ_ROOT
python utils/create_naq_dataset.py --type nlq
python utils/create_naq_dataset.py --type tacos
  • Prepare datasets for NLQ training. 
    cd $NAQ_ROOT
# Prepare NLQ dataset
python utils/prepare_ego4d_dataset.py \
    --input_train_split data/nlq_train.json \
    --input_val_split data/nlq_val.json \
    --input_test_split data/nlq_test_unannotated.json \
    --output_save_path data/dataset/nlq_official_v1

# Prepare NLQ + NaQ dataset
python utils/prepare_ego4d_dataset.py \
    --input_train_split data/nlq_aug_naq_train.json \
    --input_val_split data/nlq_val.json \
    --input_test_split data/nlq_test_unannotated.json \
    --output_save_path data/dataset/nlq_aug_naq_official_v1

# Prepare TaCOS dataset
python utils/prepare_ego4d_dataset.py \
    --input_train_split data/tacos_train.json \
    --input_val_split data/tacos_val.json \
    --input_test_split data/tacos_val.json \
    --output_save_path data/dataset/tacos_official_v1

# Prepare TaCOS + NaQ dataset
python utils/prepare_ego4d_dataset.py \
    --input_train_split data/tacos_aug_naq_train.json \
    --input_val_split data/tacos_val.json \
    --input_test_split data/tacos_val.json \
    --output_save_path data/dataset/tacos_aug_naq_official_v1