datasets

Available Datasets

Datasets can be downloaded in: https://drive.google.com/drive/folders/1AThibyepa84PKW893XV0MwZyMoKy6B_e

To use the datasets with the WSEF framework, just extract them inside the "datasets/" directory.

Currently available:

• Oxford17Flowers (https://www.robots.ox.ac.uk/~vgg/data/flowers/17/)
• Corel5k (http://www.ci.gxnu.edu.cn/cbir/Dataset.aspx)
• Cub200-2011 (http://www.vision.caltech.edu/visipedia/CUB-200-2011.html)

Note: Features were extracted using:

• Resnet152: https://github.com/Cadene/pretrained-models.pytorch
• ACC: https://github.com/dermotte/LIRE

Datasets Directory Structure

The "datasets/" directory contains a folder for each dataset, where the name must correspond to the dataset name defined in the WSEF config.py file. Inside each directory there is:

• A groundtruth.txt file that specifies the label of each class. This file uses the same format of the UDLF classes file.
• The rks directory that contains a text file for each ranked list. The files use the same format of the UDLF numeric ranked lists.
• The features directory that contains a .npz file for each feature. Each ranked list must have a corresponding feature. Each file contains a list with a feature vector for each image.

datasets
│
└───flowers
│   │   groundtruth.txt
│   │
│   └───rks
│   │   │   acc.txt
│   |   │   resnet.txt
│   |   │   ...
│   │
│   └───features
│       │   acc.npz
│       │   resnet.npz
│       │   ...
│
└───corel5k
│   │   groundtruth.txt
│   │
│   └───rks
│   │   │   acc.txt
│   |   │   resnet.txt
│   |   │   ...
│   │
│   └───features
│       │   acc.npz
│       │   resnet.npz
│       │   ...