PREPARATION.md

Preparation

Install the dependencies

pip install -r requirements-training.txt
pip install -v -e .

Data Preparation

We need to prepare ImageNet-1k datasets to do zero-shot classification task.

• ImageNet-1k

ImageNet-1k contains 1.28 M images for training and 50 K images for validation. The train set and validation set should be saved as the *.tar archives:

ImageNet/
├── train.tar
└── val.tar

Our code also supports storing images as individual files as follow:

ImageNet/
├── train
│   ├── n01440764
│   │   ├── n01440764_10026.JPEG
│   │   ├── n01440764_10027.JPEG
...
├── val
│   ├── n01440764
│   │   ├── ILSVRC2012_val_00000293.JPEG