lightly-ai/lightly-studio
Curate, Annotate, and Manage Your Data in LightlyStudio.
What it solves
LightlyStudio is an open-source tool that unifies the fragmented workflow of preparing data for computer vision models. It addresses the difficulty of managing, curating, and annotating large image and video datasets, as well as evaluating model predictions against ground truth data in a single interface.
How it works
The tool operates as a browser-based application that runs locally on the user's machine. Users load their data (from local paths or cloud storage like S3 and GCS) into a local DuckDB database using a Python interface. Once the server is started, users can explore, filter, and manage their data via the GUI. It leverages Rust for performance, allowing it to handle large datasets like COCO and ImageNet on consumer-grade hardware.
Who it’s for
It is designed for machine learning engineers and data scientists working with computer vision, specifically those who need to curate high-quality training sets, manage annotations for object detection and segmentation, and analyze model failure patterns.
Highlights
- Unified Workflow: Combines curation, annotation, and model evaluation in one tool.
- Multi-modal Support: Works with both image and video datasets.
- Advanced Sampling: Includes strategies to select representative and diverse samples to reduce labeling costs.
- Flexible Data Loading: Supports local files and major cloud storage providers (AWS S3, GCS, Azure).
- Extensible: Supports plugins for tasks like SAM autolabeling.
- Broad Format Support: Compatible with COCO and YOLO formats.