Graduate Thesis Or Dissertation

 

Semantic Image Segmentation Using Domain Constraints Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/s1784r52p

Descriptions

Attribute NameValues
Creator
Abstract
  • This dissertation addresses the problem of semantic labeling of image pixels. In the course of our work, we considered different types of semantic labels, including object classes (e.g., car, person), 3D depth values (in the range 0 to 80 meters), and affordance classes (e.g., walkable, sittable). Semantic pixel labeling is challenging as objects may appear in various poses, under partial occlusion, and against a cluttered background in the scene. To address these challenges in semantic segmentation, we developed approaches in a unified research theme that of incorporating domain knowledge in learning and inference. As our results show, domain knowledge helps to resolve various ambiguities in semantic segmentation. We addressed this problem in supervised and weakly supervised settings, where the former provides pixel-wise ground-truth annotations in training, and the latter provides ground truths only as image-level tags. Our approaches range from beam search based inference to deep convolutional neural networks (CNN). Our approaches achieved state-of-the-art performance on the benchmark datasets for all types of semantic segmentation problems.Our main contributions include:1. Efficient beam search based inference that guarantees to respect domain constraints.2. Novel deep neural architecture called neural regression forest, which integratesdecision forests with CNNs.3. Multi-scale CNN architecture for extracting and fusing diverse mid-level visualcues, including depth map, surface normals, and object localization.4. Constraint-based regularized learning of a CNN where constraints are defined asspatial relationships between objects in the domain.5. Weakly supervised learning of CNNs using neural attention cues.6. We introduced first manually annotated dataset for evaluating affordance segmentation.
License
Resource Type
DOI
Date Available
Date Copyright
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Conference Name
Conference Section/Track
Conference Location
Academic Affiliation
Non-Academic Affiliation
Keyword
Rights Statement
Publisher
Peer Reviewed
Language
Replaces
Additional Information
  • description.provenance : Submitted by Anirban Roy (royani@onid.orst.edu) on 2017-07-01T05:58:45ZNo. of bitstreams: 2license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5)RoyAnirban2017.pdf: 25019188 bytes, checksum: 16d01c8e102ecb83375639b8b898fea7 (MD5)
  • description.provenance : Approved for entry into archive by Julie Kurtz(julie.kurtz@oregonstate.edu) on 2017-07-03T16:21:02Z (GMT) No. of bitstreams: 2license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5)RoyAnirban2017.pdf: 25019188 bytes, checksum: 16d01c8e102ecb83375639b8b898fea7 (MD5)
  • description.provenance : Made available in DSpace on 2017-07-05T18:17:37Z (GMT). No. of bitstreams: 2license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5)RoyAnirban2017.pdf: 25019188 bytes, checksum: 16d01c8e102ecb83375639b8b898fea7 (MD5) Previous issue date: 2017-06-12
  • description.provenance : Approved for entry into archive by Steven Van Tuyl(steve.vantuyl@oregonstate.edu) on 2017-07-05T18:17:37Z (GMT) No. of bitstreams: 2license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5)RoyAnirban2017.pdf: 25019188 bytes, checksum: 16d01c8e102ecb83375639b8b898fea7 (MD5)
Embargo reason

Relationships

Parents:

This work has no parents.

Items