ByLabel: A Boundary Based Semi-Automatic Image Annotation Tool - PowerPoint PPT Presentation

ByLabel: A Boundary Based Semi-Automatic Image Annotation Tool Xuebin Qin, Shida He, Zichen Zhang, Masood Dehghan and Martin Jagersand Department of Computing Science University of Alberta, Canada

Introduction Our Solution Results Conclusions We are aiming at developing a boundary based semi-automatic Centriod Bounding box Quadrilateral annotation tool to acquire pixel accurate ground truth. Polygon Boundary map Region mask Fig. 1 Several types of ground truth

Introduction Our Solution Results Conclusions To acquire accurate boundary maps and region masks with light human workload, three problems have to be solved:  P.1 Many control points are  P.2 Accurately  P.3 It is hard to describe objects required to describe smoothed locating these control with holes and objects divided by and complex curves. points are difficult. occlusions. (Sampling problem) (Locating problem) (Description problem) (b) Object divided (a) Object with hole by occlusions Fig. 2 Illustration of the three problems in annotation

Introduction Our Solution Results Conclusions Our annotation tool has the following workflow: Fig. 3 Workflow of our method

Introduction Our Solution Results Conclusions Feature Detection Detect Edge Segments and Split them into Edge Fragments ( Solves locating pro. ) (b) Edge Fragments (a) Input image Boundaries Manually select fragments Labeling to form closed boundaries Objects (Solves sampling pro.) Annotation Group boundaries belong to the same object and input the class name (Solves description pro.) (d) Output region mask (c) Labeled boundaries Fig. 4 Annotate an object step by step

Introduction Our Solution Results Conclusions A1 A2 A3 A4 A5 B1 B2 B3 B4 B5 C1 C2 C3 C4 C5 Fig. 5 15 Testing images in three groups

Introduction Our Solution Results Conclusions Table. 1. Average Clicks Table. 2. Average Time Costs (s) Table. 3. Average Error (pixel)

Introduction Our Solution Results Conclusions (b) Earphone (a) Pedestrian (c) Bicycle (d) Input image (e) Classes annotation (f) Instances annotation Fig. 6 Typical annotation results

Introduction Our Solution Results Conclusions ByLabel greatly improves the annotation efficiency and accuracy . It also reduces the annotation uncertainty and error .

Questions ??? Poster: 3C-9 ByLabel http://webdocs.cs.ualberta.ca/~vis/bylabel/ xuebin@ualberta.ca, mj7@ualberta.ca