Dataset Format

Description of Dataset Folder Structures

Before you start uploading your datasets to our platform, it’s important to understand the folder structures for the various dataset formats we support. Each format—ImageNet, LabelBox, and MSCOCO—has its own unique way of organizing data, which is essential for ensuring compatibility with our platform’s tools and workflows.

ImageNet Format

The ImageNet format is typically organized into a hierarchical structure that reflects the categories and subcategories of the dataset. This format is particularly useful for large-scale image classification tasks.

Folder Structure:

ImageNet/
│
├── train/
│   ├── class1/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   ├── class2/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   └── ...
│
├── val/
│   ├── class1/
│   │   ├── image1.jpg
│   │   └── ...
│   ├── class2/
│   │   ├── image1.jpg
│   │   └── ...
│   └── ...
│
└── test/
    ├── class1/
    │   ├── image1.jpg
    │   └── ...
    ├── class2/
    │   ├── image1.jpg
    │   └── ...
    └── ...

Each image file is typically named uniquely within its class folder.

LabelBox Format

LabelBox datasets are often used for supervised learning tasks, especially in computer vision, where annotated images are required. The LabelBox format typically includes JSON files containing the annotations and image files.

Folder Structure

LabelBox/
│
├── images/
│   ├── image1.jpg
│   ├── image2.jpg
│   └── ...
│
├── annotations/
│   ├── image1.json
│   ├── image2.json
│   └── ...
│
└── data.json  (optional)

images: Contains all the images in the dataset. annotations: Contains individual JSON files where each file corresponds to the annotations for a single image. data.json (optional): Sometimes, a single consolidated JSON file is provided that contains all the annotations for the dataset.

Each annotation file includes metadata about the image and the annotated objects, such as bounding boxes, segmentation masks, or class labels.

MSCOCO Format

The MSCOCO (Microsoft Common Objects in Context) format is widely used in object detection, segmentation, and keypoint detection tasks. The structure is designed to accommodate complex annotations, including bounding boxes, segmentation masks, and object categories.

Folder Structure:

MSCOCO/
│
├── images/
│   ├── train/
│   │   ├── 000000000001.jpg
│   │   ├── 000000000002.jpg
│   │   └── ...
│   ├── val/
│   │   ├── 000000000001.jpg
│   │   ├── 000000000002.jpg
│   │   └── ...
│   └── test/
│       ├── 000000000001.jpg
│       ├── 000000000002.jpg
│       └── ...
│
└── annotations/
    ├── instances_train.json
    ├── instances_val.json
    └── instances_test.json

  • images: Contains subfolders for training, validation, and test images.

    • train : Contains all training images.

    • val : Contains all validation images.

    • test : Contains all test images.

  • annotations : Contains JSON files for annotations.

    • instances_train.json: Contains instance-level annotations (e.g., bounding boxes, segmentation masks) for the training images.

    • instances_val.json: Contains instance-level annotations for the validation images.

    • instances_test.json: Contains instance-level annotations for the test images.

This format is highly structured and supports complex datasets with multiple annotation types.

Conclusion

By organizing your datasets according to these structures, you’ll ensure compatibility with our tools and facilitate seamless integration into your workflow. Whether you’re working with ImageNet, LabelBox, or MSCOCO formats, our platform provides the flexibility and power you need to manage your datasets effectively.