Label Studio: set up local storage for image annotation

This tutorial configures local file storage in the open source edition of Label Studio on Windows. The presenter installs it in a separate Conda environment with Python 3.10, citing compatibility issues with newer Python versions. Linux and Ubuntu users are directed to the documentation for command differences.

The setup enables local file serving and sets a document root above the dataset folders. Separate input and output directories hold source images and annotation results. The presenter uses Windows setx commands for persistent environment variables, then connects the folders through the project's Cloud Storage settings with Local Files as the provider.

The image import walkthrough includes two useful corrections. Scanning subfolders lets Label Studio find nested images without moving them into the top directory. When synchronization fails because the import method expects JSON, the presenter switches to creating a task for each storage object. The example uses roughly a thousand images; it does not establish an unlimited capacity.

The final section configures a bounding-box labeling template with a car label, submits sample annotations and exports selected tasks as JSON. It also checks the output directory and explains the target storage option that controls whether deleting tasks removes their stored annotations.