Annotation Editing Tool

Overview

The annotation editing tool was envisioned as an interactive toolbox for editing LIDAR datasets as an extension of the LIDAR Visualization Tool. Extending the modular functionality of LVT for generic datasets, Annotation Editing Tool allows researchers to edit any supported datatypes in real time through a variety of tools. After conversion of a datatype into the LVT generic format, a user can open the editing tool at any time during viewing allowing users to interact with the current frame and scene from that dataset. Within the editing mode, not only can users add new bounding boxes, they can modify and delete preexisting boxes placed there by predictive algorithms. Tools are intuitive and enable a wide range of control over an annotations position, rotation, size and annotation class. Through this fine-tuned editing, researchers can use the Annotation Editing Tool to improve their predictive models by more accurately labeling objects in scene and removing false positives. Finally, the tool allows users to convert their data back into the appropriate proprietary datatype for use in their predictive models.

Background

This project was a collaboration between 7 undergraduate students at University of Maryland during the Fall 2022 semester as part of a semester long assignment for CMSC 435 under the guidance of both professor James Purtilo and a graduate fellow at Carnegie Mellon University.