Semantic amodal video segmentation using a synthetic dataset

In this work, we provide tools for annotating both object category and shot transitions for a new semantic modal instance-level object segmentation dataset. This new dataset provides ample opportunities to train models for instance-level segmentation, both modal and amodal. Moreover, in this work, we also present results for instance-level segmentation using ResNet-based DeepLab, a state-of-the-art semantic image segmentation model. We also develop a new semantic amodal instance-level video segmentation model based on DeepLab for the aforementioned dataset. Our model for amodal segmentation operates on a per-frame basis, and the model is guided by the modal mask estimated from the current frame and from previous frames delineating the object of interest. We demonstrate the efficacy of the proposed model on the new dataset.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Hui, Kexin

Contributors dc:contributor

• Schwing, Alexander

• ANNOTATION
• VIDEO SEGMENTATION

Statement dc:rights

• Copyright 2018 Kexin Hui

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/102966

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/102966