UCP-Net: Unstructured Contour Points for Instance Segmentation
Camille Dupont∗, Yanis Ouakrim∗ and Quoc Cuong Pham Université Paris-Saclay, CEA, List, F-91120, Palaiseau, France. Email: {camille.dupont, *Equal contribution
Abstract
The goal of interactive segmentation is to assist users in producing segmentation masks as fast and as accurately as possible. Interactions have to be simple and intuitive and the number of interactions required to produce a satisfactory segmentation mask should be as low as possible. In this paper, we propose a novel approach to interactive segmentation based on unconstrained contour clicks for initial segmentation and segmentation refinement. Our method is class-agnostic and produces accurate segmentation masks (IoU $>$ 85%) for a lower number of user interactions than state-of-the-art methods on popular segmentation datasets (COCO MVal, SBD and Berkeley).
中文速览
用鼠标在目标物体轮廓上随意点几下就能完成精准分割——这正是本文要实现的目标。现有方法要么需要用户精确点击物体的上下左右四个极端点（认知负担重、固定4次点击不够灵活），要么依赖大量正负样本点，效率偏低。本文提出了一种基于"无约束轮廓点击（unconstrained contour clicks）"的交互式分割方法：用户只需在目标轮廓上随意点2到多个点，系统便自动确定感兴趣区域并通过深度网络预测分割掩码，不满意时可继续补充点击迭代修正。在COCO MVal、SBD、Berkeley等主流数据集上，该方法以更少的交互次数就达到了85%以上的IoU，超越现有最优方法，且模型可在浏览器端实时运行（无需独立GPU，128×128实例仅需53毫秒），对降低图像标注成本、推动深度学习数据生产具有实际意义。
原文 arXiv:2109.07592;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2109.07592v1