Iterative Bounding Box Annotation for Object Detection

TL;DR

This paper introduces an iterative semi-automatic bounding box annotation method that significantly reduces manual effort in object detection labeling by training on small batches and learning to propose bounding boxes for correction.

Contribution

It presents a novel iterative approach that combines human correction with machine learning to improve annotation efficiency in object detection tasks.

Findings

Reduces manual annotation effort by up to 75%.

Effective across multiple datasets.

Optimizes data presentation order for better efficiency.

Abstract

Manual annotation of bounding boxes for object detection in digital images is tedious, and time and resource consuming. In this paper, we propose a semi-automatic method for efficient bounding box annotation. The method trains the object detector iteratively on small batches of labeled images and learns to propose bounding boxes for the next batch, after which the human annotator only needs to correct possible errors. We propose an experimental setup for simulating the human actions and use it for comparing different iteration strategies, such as the order in which the data is presented to the annotator. We experiment on our method with three datasets and show that it can reduce the human annotation effort significantly, saving up to 75% of total manual annotation work.