Skip to content

Overview

Definitions

Supervisely project is a way of organizing annotated images.

A project is a named set of datasets. A dataset is a named set of annotated images. Each image has corresponding annotation, and vice versa.

Dataset name is unique within a project, and image name is unique within a dataset.

Annotation

Please note that in terms of projects "annotation" means full set of objects (figures) linked with some image.

Project in filesystem

When stored on disk, project has the following structure:

my_project
├── meta.json
├── dataset_name_01
│   ├── ann
│   │   ├── my_favorite_image.json
│   │   ├── img_y.json
│   │   └── img_z.json
│   └── img
│       ├── my_favorite_image.jpg
│       ├── img_y.png
│       └── img_z.jpeg
└── dataset_name_02
    ├── ann
    │   ├── img_x.json
    │   ├── img_y.json
    │   └── img_z.json
    └── img
        ├── img_x.jpg
        ├── img_y.jpg
        └── img_z.jpg

In details:

  • The project is called "my_project".

  • Project directory ("my_project" in the example) contains some dataset directories and file meta.json.

  • meta.json contains metainfo about project: figure classes and image tags.

  • Each dataset directory contains two directories, ann with JSON annotations and img with images.

  • Each image shares name (e.g., img_x) with corresponding anotation.

  • PNGs and JPGs images are supported.

How to use

Look at simple examples:

Create project structure with ProjectStructure or read it from disk with ProjectFS.

Work with project meta using ProjectMeta.

Work with single annotation using Annotation.