EVA is a web-based tool for efficient annotation of videos and image sequences with built-in tracking capabilities.
The tool addresses the need to label objects in video and image data at the bounding box level. It provides a browser-based interface where users can annotate frames, and includes a tracking component that can follow objects across frames. Annotated data can be exported in YOLO or Pascal VOC format, making it compatible with common computer vision workflows.
EVA suits teams that need to prepare labeled video datasets for object detection or tracking tasks. The tool runs locally through a web interface accessible in Chrome or Firefox, and supports both single images and video sequences. It is positioned as a redesign of an earlier annotation tool with the addition of tracking capabilities. Setup is straightforward on Windows, Linux, and Docker, though Windows users must manually place ffmpeg in the tool directory. The tracker parameters can be customized by editing the configuration file.
The project accepts external contributions and acknowledges specific contributors in its documentation. Development appears focused on maintaining the core annotation and tracking functionality rather than rapid feature expansion.
