Ulmer, Alex (2015)
Joint Estimation of Depth and Labels from a Single Image.
Technische Universität Darmstadt
Masterarbeit, Bibliographie
Kurzbeschreibung (Abstract)
Estimating depth and semantic segmentation from a single image are two very challenging tasks in computer vision. In traditional approaches models for image structure or semantic class connections were used to create estimations. In the past years new non-parametric methods proved to achieve state of the art results for each of the problems. The non-parametric approach makes use of huge image databases by directly transferring the ground truth to the query image. Further, research showed that semantic segmentation can be used to improve depth estimation and vice versa. In this work both problems of depth estimation and semantic segmentation are tackled. Therefore a non-parametric system is build which extracts features from a query images, retrieves similar images, matches sub regions of the images, and generates a pixel-level potential for depth and semantic segmentation. This potential is utilized for a joint optimization with a fully connected conditional random field to achieve consistent estimates. The system is tested with synthetic images which have pixel accurate depth maps and semantic segmentation. Additionally, an evaluation on publicly available real world datasets shows that the system achieves competitive performance.
| Typ des Eintrags: | Masterarbeit |
|---|---|
| Erschienen: | 2015 |
| Autor(en): | Ulmer, Alex |
| Art des Eintrags: | Bibliographie |
| Titel: | Joint Estimation of Depth and Labels from a Single Image |
| Sprache: | Englisch |
| Publikationsjahr: | 2015 |
| Kurzbeschreibung (Abstract): | Estimating depth and semantic segmentation from a single image are two very challenging tasks in computer vision. In traditional approaches models for image structure or semantic class connections were used to create estimations. In the past years new non-parametric methods proved to achieve state of the art results for each of the problems. The non-parametric approach makes use of huge image databases by directly transferring the ground truth to the query image. Further, research showed that semantic segmentation can be used to improve depth estimation and vice versa. In this work both problems of depth estimation and semantic segmentation are tackled. Therefore a non-parametric system is build which extracts features from a query images, retrieves similar images, matches sub regions of the images, and generates a pixel-level potential for depth and semantic segmentation. This potential is utilized for a joint optimization with a fully connected conditional random field to achieve consistent estimates. The system is tested with synthetic images which have pixel accurate depth maps and semantic segmentation. Additionally, an evaluation on publicly available real world datasets shows that the system achieves competitive performance. |
| Freie Schlagworte: | Patch-based depth reconstruction, Depth images, Semantic labeling, Inference |
| Zusätzliche Informationen: | 92 p. |
| Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Graphisch-Interaktive Systeme |
| Hinterlegungsdatum: | 10 Mai 2019 06:26 |
| Letzte Änderung: | 10 Mai 2019 06:26 |
| PPN: | |
| Export: | |
| Suche nach Titel in: | TUfind oder in Google |
 |
Frage zum Eintrag |
Optionen (nur für Redakteure)
 |
Redaktionelle Details anzeigen |
Drucken
Impressum