DATASETS.md

March 4, 2026 · View on GitHub

Dataset Preparation

  1. Download all 11 prompt-tuning datasets with foreground views on [🤗HuggingFace].
    (This dataset is fully compatible with other prompt tuning methods that do not use foreground views!)
    If you need a separate mask data that does not contain raw images, please download it from the following links: Mask-Only-Data

  2. 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 should be organized as:

    • ImageNet: (Please first prepare raw ImageNet dataset from: [Raw Images] [annoations] [val conversion script] )

      $DATA/
|–– imagenet-1k/
|   |–– imagenet/
|   |   |-- images/
|   |   |   |-- train/
|   |   |   |   |-- n01440764/
|   |   |   |   |-- n01443537/
|   |   |   |   |-- ...
|   |   |   |-- val/
|   |   |   |   |-- n01440764/
|   |   |   |   |-- n01443537/
|   |   |   |   |-- ...
|   |   |   |-- test/
|   |   |   |   |-- ILSVRC2012_test_00000001.JPEG
|   |   |   |   |-- ILSVRC2012_test_00000002.JPEG
|   |   |   |   |-- ...
|   |   |–– classnames.txt
|   |   |–– command.txt
|   |–– mask/

    • Caltech101:

      $DATA/
|–– caltech-101/
|   |–– caltech-101/
|   |   |-- 101_ObjectCategories/
|   |   |-- Annoations/
|   |   |-- ...
|   |–– mask/

    • Food101:

       $DATA/
 |–– food101/
 |   |–– food-101/
 |   |   |-- images/
 |   |   |-- meta/
 |   |   |-- ...
 |   |–– mask/

    • StanfordCars:

       $DATA/
 |–– stanford_cars/
 |   |–– stanford_cars/
 |   |   |-- cars_train/
 |   |   |-- cars_test/
 |   |   |-- ...
 |   |–– mask/

    • OxfordPets:

      $DATA/
|–– oxford_pets/
|   |–– oxford_pets/
|   |   |-- annoations/
|   |   |-- images/
|   |   |-- ...
|   |–– mask/

    • Flowers102:

      $DATA/
|–– flowers-102/
|   |–– oxford_flowers/
|   |   |-- jpg/
|   |   |-- cat_to_name.json
|   |   |-- ...
|   |–– mask/

    • DTD:

      $DATA/
|–– dtd/
|   |––dtd/
|   |   |-- dtd/
|   |   |   |-- images/
|   |   |   |-- imdb/
|   |   |   |-- ...
|   |   |–– mask/

    • EuroSAT:

      $DATA/
|–– eurosat/
|   |–– eurosat/
|   |   |-- 2750/
|   |   |-- split_zhou_EuroSAT.json
|   |–– mask/

    • FGVC_Aircraft:

      $DATA/
|–– fgvc_aircraft/
|   |–– fgvc-aircraft-2013b/
|   |   |-- fgvc_aircraft
|   |   |   |-- images/
|   |   |   |-- families.txt
|   |   |   |-- ...
|   |   |–– mask/
|   |   |–– ...

    • SUN397:

       $DATA/
 |–– sun-397/
 |   |–– sun397/
 |   |   |-- sun397/
 |   |   |   |-- a/
 |   |   |   |-- b/
 |   |   |   |-- ...
 |   |   |-- classnames.txt
 |   |   |-- split_zhou_SUN397.json
 |   |   |-- ...
 |   |–– mask/

    • UCF101:

        $DATA/
  |–– ucf101/
  |   |–– ucf101/
  |   |   |-- UCF-101-midframes/
  |   |   |-- ucfTrainTestlist/
  |   |   |-- split_zhou_UCF101.json
  |   |–– mask/