|DIAS, PHILIPE - Marquette University
|SHEN, ZHOU - Marquette University
|MEDEIROS, HENRY - Marquette University
Submitted to: IEEE Winter Conference on Applications of Computer Vision
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/5/2018
Publication Date: 3/7/2019
Citation: Dias, P.A., Shen, Z., Tabb, A., Medeiros, H. 2019. FreeLabel: A publicly available annotation tool based on freehand traces. IEEE Winter Conference on Applications of Computer Vision. https://doi.org/10.1109/WACV.2019.00010.
Interpretive Summary: We examined the problem of marking (also referred to as annotating) images. In our context, the marking needs to be done at the pixel level, which means that workers need to color in regions of digital images that match certain objects, such as car, flower, bicycle, etc. This work is tedious and detailed, but is necessary to train algorithms and models to automatically recognize these object categories. We created a tool to help workers annotate images more rapidly, and analyzed worker behavior with our tool. The impact of this work is that well-annotated datasets can be produced more rapidly, which has implications for producing better recognition models in agriculture and other relatively dataset-poor domains.
Technical Abstract: Large-scale annotation of image segmentation datasets is often prohibitively expensive, as it usually requires a huge number of worker hours to obtain high-quality results. Abundant and reliable data has been, however, crucial for the advances on image understanding tasks recently achieved by deep learning models. In this paper, we introduce FreeLabel, an intuitive open-source web interface that allows users to obtain high-quality segmentation masks with just a few freehand scribbles, in a matter of seconds. The efficacy of FreeLabel is quantitatively demonstrated by experimental results on the PASCAL dataset as well as on a dataset from the agricultural domain. Designed to benefit the computer vision community, FreeLabel can be used for both crowdsourced or private annotation and has a modular structure that can be easily adapted for any image dataset.