preparing_datasets.md

Preparing Datasets

ImageNet and CIFAR datasets

• Download the original train and test set for ImageNet & ImageNet-C datasets.
• Download the original train and test set for CIFAR-10 & CIFAR-10C datasets.

KITTI dataset

• Download Clear (Original) KITTI dataset.
• Download KITTI-Fog/Rain datasets.
• Super-impose snow on KITTI dataset through this repository.
• Generate labels YOLO can use (see Dataset directory structures subsection).

To generate labels YOLO can use from the original KITTI labels run

python main.py --kitti_to_yolo_labels /path/to/original/kitti

This is expecting the path to the original KITTI directory structure

path_to_specify
└── raw
    └── training
        ├── image_2
        └── label_2

Which will create a yolo_style_labels directory in the raw directory, containing the KITTI labels in a format YOLO can use.

Structure the choosen dataset(s) as described here.