This repo contains scripts for downloading datasets and knowledge files of all tasks in ELEVATER. In this repo, we provide a central hub to all the datasets along with our collected knowledge files to facilitate future end-to-end training on ELEVATER benchmark. Note that:
Note: The data is on Azure Storage Blob, a SAS with Read permission is provided. Please append the following SAS at the end of each link to download:
?sp=r&st=2023-08-28T01:41:20Z&se=3023-08-28T09:41:20Z&sv=2022-11-02&sr=c&sig=Msoq5dIl%2Fve6F01edGr8jgcZUt7rtsuJ896xvstSNfM%3D
Install
pip install vision-datasets==0.2.17
cd classification
Download all datasets
python download_datasets.py --ds all
Download a specific dataset
python download_datasets.py --ds `DATASET_NAME`
# `DATASET_NAME` should be one of ['caltech-101', 'cifar-10', 'cifar-100', 'country211', 'dtd', 'eurosat_clip', 'fer-2013', 'fgvc-aircraft-2013b-variants102', 'food-101', 'gtsrb', 'hateful-memes', 'kitti-distance', 'mnist', 'oxford-flower-102', 'oxford-iiit-pets', 'patch-camelyon', 'ping-attack-on-titan-plus', 'ping-whiskey-plus', 'rendered-sst2', 'resisc45_clip', 'stanford-cars', 'voc-2007-classification']
Download knowledge files
python download_knowledge.py
cd detection
Download all datasets
python download.py
Download a specific dataset
python download.py --dataset_names `DATASET_NAME`
Knowledge files are already included in the repo
detection/*yaml