VIAME/VIAME
Video and Image Analytics for Multiple Environments
What it solves
VIAME is a comprehensive computer vision toolkit designed for "do-it-yourself" AI, enabling users to perform object detection, tracking, and image/video annotation without needing deep expertise in AI development. Originally created for marine species analytics, it now serves as a generic toolkit for various computer vision tasks.
How it works
The system is built on a multi-threaded pipeline framework (Kwiver) that allows the integration of nodes written in C++, Python, and MATLAB. It provides a suite of standalone tools, plugins for various AI frameworks (such as PyTorch, TensorFlow, TensorRT, and Darknet/YOLO), and user interfaces (DIVE, VIEW, and SEARCH) for annotation, visualization, and model generation.
Who it’s for
It is intended for researchers and practitioners who need to process video data, annotate images, and generate AI models for object detection and tracking, particularly those who may not be professional software engineers or AI researchers.
Highlights
- Multi-Framework Support: Integrates detectors and trackers from PyTorch, TensorFlow, TensorRT, and Darknet.
- Extensible Pipeline: A multi-threaded framework that connects diverse language nodes (C++, Python, MATLAB).
- Integrated Tooling: Includes tools for image mosaicing, size measurement, image enhancement, and algorithm evaluation.
- Flexible Deployment: Available as desktop binaries for Windows and Linux, Docker images, and a web-based annotator.