Date of Award
6-21-2019
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Computer Science
First Advisor
Mariofanna Milanova
Abstract
With the advent of object detection and instance segmentation algorithms, the complexity of annotating training data has compounded. In traditional image classification tasks, training images must only be given a label. This type of training data alone is not enough to perform detection or instance segmentation as we need additional object level information like bounding boxes or segmentation masks appended to the annotation. There are many tools that help a user process images to produce data annotations that contain bounding box and segmentation mask information. However, most of these are semi-supervised and require a user to be present for each annotation creation. This study’s primary goal is to introduce a modified Mask R CNN model that will reduce image annotation time while maintaining a mean average precision (mAP) similar to other studies. The results show that it is possible to expand a training set of segmentation masks while minimizing the reduction in quality. While there are dataset considerations, such as diversity of object orientation and class similarity, we show that, given appropriate preprocessing and dataset selection, there are use cases for our data annotation approach.
Recommended Citation
Miller, Michael, "Incremental Data Annotation with Mask R-CNN" (2019). Theses and Dissertations. 883.
https://research.ualr.edu/etd/883
