Overview

Import data in Cityscapes format to Supervisely from a folder or archive.

Preparation

Upload your data as a project folder in Cityscapes format to Team Files.
ℹ️ You can download the archive with data example here.

The structure of the project has to be the following:

📦project folder or archive
├──📂gtFine
│   ├──📂test
│   │   └──...
│   ├──📂train
│   │   └──...
│   └──📂val
│       └──...
└──📂leftImg8bit
    ├──📂test
    │   └──...
    ├──📂train
    │   └──...
    └──📂val
        └──...

⚠️ Note: Import will crash if the archive with the parent directory doesn't contain corresponding data.

How To Run

Step 1: Add the app to your team from Ecosystem if it is not there.

Step 2: Go to the Team Files page, right-click on your archive or folder, containing cityscapes data, and choose Run App->Import Cityscapes.

Another way to import is to find the app page in Ecosystem and click RUN APP... button. Then you will see a modal window.

There you can drag-and-drop projects or choose from Team Files too.

Step 3: Set target slider value to split image sets to train and val and assign corresponding tags(only for cases, when the image set doesn't contain val folder). Press the RUN button. As a result, you will be redirected to Workspace Tasks page.

How to use

The resulting project will be placed in your current Workspace with name in format {archive/folder name}. Images in datasets will have tags (train, val, or test) corresponding to the parent directories in which the datasets were located during import. If input data(archive or folder) contains train folder and doesn't contain val folder, the images from train folder will be tagged with train and val tags according to the exposed slider ratio.

You can also access your project by clicking on its name from the Workspace Tasks page.