Skip to content

Latest commit

 

History

History
315 lines (248 loc) · 11.6 KB

DATASETS.md

File metadata and controls

315 lines (248 loc) · 11.6 KB
Raw

How to install downstream datasets

The dataset installation instruction is modified from official CoOp repository, with some references of updates from CMLP.

We suggest putting all datasets under the same folder (say $DATA) to ease management and following the instructions below to organize datasets to avoid modifying the source code. The file structure looks like:

$DATA/
|–– semi-aves/
|–– flowers102/
|–– fgvc-aircraft/
|–– eurosat/
|–– dtd/
|–– oxford_pets/
|–– stanford_cars/
|–– food101/
|–– imagenet/

$RETRIEVED/
|–– semi-aves/
|–– flowers102/
|–– fgvc-aircraft/
|–– eurosat/
|–– dtd/
|–– oxford_pets/
|–– stanford_cars/
|–– food101/
|–– imagenet/

Update the config.yml with the path to the datasets and retrieved data.

dataset_path: /scratch/group/real-fs/dataset/
retrieved_path: /scratch/group/real-fs/retrieved/

If you have some datasets already installed somewhere else, you can create symbolic links in $DATA/dataset_name that point to the original data to avoid duplicate download.

Below we provide instructions to prepare the downstream datasets used in our experiments. Please refer to RETRIEVAL.md for instructions on how to setting up the retrieved datasets.

Datasets list:

The instructions to prepare each dataset are detailed below. To ensure reproducibility and fair comparison for future work, we provide fixed train/val/test splits for all datasets except ImageNet where the validation set is used as test set. The fixed splits are either from the original datasets (if available) or created by us. For few-shot training data, we sample from given training data with 3 random seeds. The splits are available in the SWAT/data/{dataset}/ folder.

Semi-Aves

  • Create a folder named semi-aves/ under $DATA.
  • Download data from the official repository or following the wget commands below
cd $DATA/semi-aves/

# train_val data
wget https://drive.google.com/uc?id=1xsgOcEWKG9CszNNT_EXN3YB1OLPYNbf8 

# test
wget https://drive.google.com/uc?id=1OVEA2lNJnYM5zxh3W_o_Q6K5lsNmJ9Hy

# unlabeled_ID
wget https://drive.google.com/uc?id=1BiEkIp8yuqB5Vau_ZAFAhwp8CNoqVjRk

# unzip
tar -xzf *.gz
  • The annotations are extracted from the official annotation json files. We have reformatted labels and provided to you as ltrain.txt, ltrain+val.txt,val.txt and test.txt in the SWAT/data/semi-aves/ folder.

The directory structure should look like:

semi-aves/
|–– trainval_images
|–– u_train_in
|–– test

Flowers102

wget https://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz
  • We have reformatted labels and provided to you as train.txt, val.txt and test.txt in the SWAT/data/flowers102/ folder.

The directory structure should look like:

flowers102/
|–– jpg
|–– imagelabels.mat
|–– cat_to_name.json

FGVC-Aircraft

wget https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz
  • Extract fgvc-aircraft-2013b.tar.gz and keep only data/.
  • Move fgvc-aircraft-2013b/data/ to $DATA/fgvc-aircraft.
  • We have reformatted labels and provided to you as train.txt, val.txt and test.txt in the SWAT/data/fgvc-aircraft/ folder.

The directory structure should look like:

fgvc_aircraft/
|–– fgvc-aircraft-2013b/
    |–– data/

EuroSAT

wget https://zenodo.org/records/7711810/files/EuroSAT_RGB.zip?download=1
  • Renmae the extracted folder 2750 to EuroSAT_RGB.
  • We have reformatted labels and provided to you as train.txt, val.txt and test.txt in the SWAT/data/eurosat/ folder.

The directory structure should look like:

eurosat/
|–– EuroSAT_RGB/

DTD

wget https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz
  • We have reformatted labels and provided to you as train.txt, val.txt and test.txt in the SWAT/data/dtd/ folder.

The directory structure should look like:

dtd/
|–– dtd/
    |–– images/
    |–– imdb/
    |–– labels/

OxfordPets

The directory structure should look like:

oxford_pets/
|–– images/
|–– annotations/
|–– split_zhou_OxfordPets.json

StanfordCars

  • Create a folder named stanford_cars/ under $DATA.
  • In case the following link breaks, download dataset from Kaggle.
  • Download car_devkit.tgz
wget https://github.com/pytorch/vision/files/11644847/car_devkit.tgz
tar -xzvf car_devkit.tgz

The directory structure should look like

stanford_cars/
|–– cars_test\
|–– cars_annos.mat
|–– cars_train\
|–– split_zhou_StanfordCars.json

Food101

wget http://data.vision.ee.ethz.ch/cvl/food-101.tar.gz
  • Download split_zhou_Food101.json from here.

The directory structure should look like

food-101/
|–– images/
|–– license_agreement.txt
|–– meta/
|–– README.txt
|–– split_zhou_Food101.json

Caltech101

  • Create a folder named caltech-101/ under $DATA.
  • Download 101_ObjectCategories.tar.gz from https://data.caltech.edu/records/mzrjq-6wc02 and extract the file under $DATA/caltech-101.
  • Download split_zhou_Caltech101.json from this link and put it under $DATA/caltech-101.

The directory structure should look like

caltech-101/
|–– 101_ObjectCategories/
|–– Annotations/
|–– split_zhou_Caltech101.json

SUN397

The directory structure should look like

sun397/
|–– SUN397/
|–– split_zhou_SUN397.json
|–– ... # a bunch of .txt files

UCF101

  • Create a folder named ucf101/ under $DATA.
  • Download the zip file UCF-101-midframes.zip from here and extract it to $DATA/ucf101/. This zip file contains the extracted middle video frames.
  • Download split_zhou_UCF101.json from this link.

The directory structure should look like

ucf101/
|–– UCF-101-midframes/
|–– split_zhou_UCF101.json

ImageNet

  • Create a folder named imagenet/ under $DATA.
  • Download split_ImageNet.json (google drive link) to this folder. (Note that the original CoOp does not make a train/val split.)
  • Create images/ under imagenet/. -Option 1: download dataset from Huggingface repo
# create a HF account and get the API token, then run the following command
wget --header="Authorization: Bearer YOUR_HUGGINGFACE_TOKEN" https://huggingface.co/datasets/ILSVRC/imagenet-1k/resolve/main/data/train_images_0.tar.gz

mkdir train
tar -xzf train_images_0.tar.gz -C train/

mkdir val
tar -xzf val_images.tar.gz -C val/
  • Option 2: Download the dataset from the official website and extract the training and validation sets to $DATA/imagenet/images.
  • Download the classnames.txt to $DATA/imagenet/ from this link. The class names are copied from CLIP.

The directory structure should look like

imagenet/
|–– classnames.txt
|–– images/
|   |–– train/ # contains 1,000 folders like n01440764, n01443537, etc.
|   |–– val/
  • If you had downloaded the ImageNet dataset before, you can create symbolic links to map the training and validation sets to $DATA/imagenet/images.
  • Download the classnames.txt to $DATA/imagenet/ from this link. The class names are copied from CLIP.

Prepare the dataset labels

  • We have already prepared the train.txt, val.txt and test.txt files for each dataset, using the same splits as in CoOp repository.
  • In case you are interested, we use the following script to prepare the dataset labels.
cd SWAT/
python prepare_datasets_labels.py

Prepare the few-shot labels

  • Prepare the few-shot annotation files for each dataset, using 3 random seeds.
  • We have already prepared the fewshot{4/8/16}_seed{1/2/3}.txt files for each dataset in SWAT/data/{dataset}/ folder.
python prepare_fewshot_txt.py