Author: QIU Tian
Affiliation: Zhejiang University
English | 简体中文
The default --data_root
is ./data
. Put your datasets in this directory.
Each dataset is stored in a separate directory. For example,
data/
├── mnist/
├── cifar10/
├── cifar100/
├── imagenet1k/
├── imagenet21k/
├── stl10/
├── svhn/
├── pets/
├── flowers/
├── cars/
└── your_dataset/
Note that directory names are all lowercase.
If you want to change the data root, please set the argument --data_root
.
The folder
format dataset is commonly used in image classification and can be loaded by ImageFolder
implemented
in qtcls/datasets/folder.py
. ImageNet
is of the typical folder
format. The dataset
folder/directory contains the split subfolders (train
, val
, etc.). For each split, images belonging to a particular
category are stored in a subfolder with the same name as the category. Here is an example:
dataset_name/
├── train/
├── category1/
├── image1
├── image2
├── image3
├── ...
├── category2/
├── image1
├── image2
├── image3
├── ...
├── category3/
├── .../
├── val/
├── category1/
├── image1
├── image2
├── image3
├── ...
├── category2/
├── image1
├── image2
├── image3
├── ...
├── category3/
├── .../
├── .../