Interactive Segmentation for Diverse Gesture Types Without Context

Interactive segmentation entails a human marking an image to guide how a model either creates or edits a segmentation. Our work addresses limitations of existing methods: they either only support one gesture type for marking an image (e.g., either clicks or scribbles) or require knowledge of the gesture type being employed, and require specifying whether marked regions should be included versus excluded in the final segmentation. We instead propose a simplified interactive segmentation task where a user only must mark an image, where the input can be of any gesture type without specifying the gesture type. We support this new task by introducing the first interactive segmentation dataset with multiple gesture types as well as a new evaluation metric capable of holistically evaluating interactive segmentation algorithms. We then analyze numerous interactive segmentation algorithms, including ones adapted for our novel task. While we observe promising performance overall, we also highlight areas for future improvement. To facilitate further extensions of this work, we publicly share our new dataset at https://github.com/joshmyersdean/dig.

1. Introduction

A common goal is to locate regions, such as objects or object parts, in images. We refer to this task as region segmentation. A challenge is that fully-automated solutions often are error-prone while exclusive reliance on human annotations is costly and time-consuming. As a middle ground, interactive segmentation methods empower humans to supply minimal input towards collecting consistently high-quality region segmentations. Two popular interactive segmentation settings are to generate a segmentation (1) when the only input is a human marking on an image (i.e., segmentation creation) and (2) when the input is a human marking and a previous segmentation (i.e., segmentation refinement).