<> "The repository administrator has not yet configured an RDF license."^^ . <> . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms"^^ . "This thesis investigates performance metrics and test datasets used for the evaluation of depth estimation algorithms.\r\n\r\nStereo and light field algorithms take structured camera images as input to reconstruct a depth map of the depicted scene. Such depth estimation algorithms are employed in a multitude of practical applications such as industrial inspection and the movie industry. Recently, they have also been used for safety-relevant applications such as driver assistance and computer assisted surgery.\r\nDespite this increasing practical relevance, depth estimation algorithms are still evaluated with simple error measures and on small academic datasets. To develop and select suitable and safe algorithms, it is essential to gain a thorough understanding of their respective strengths and weaknesses.\r\n\r\nIn this thesis, I demonstrate that computing average pixel errors of depth estimation algorithms is not sufficient for a thorough and reliable performance analysis. The analysis must also take into account the specific requirements of the given applications as well as the characteristics of the available test data.\r\n\r\nI propose metrics to explicitly quantify depth estimation results at continuous surfaces, depth discontinuities, and fine structures. These geometric entities are particularly relevant for many applications and challenging for algorithms. In contrast to prevalent metrics, the proposed metrics take into account that pixels are neither spatially independent within an image nor uniformly challenging nor equally relevant. \r\n\r\nApart from performance metrics, test datasets play an important role for evaluation. Their availability is typically limited in quantity, quality, and diversity. I show how test data deficiencies can be overcome by using specific metrics, additional annotations, and stratified test data.\r\n\r\nUsing systematic test cases, a user study, and a comprehensive case study, I demonstrate that the proposed metrics, test datasets, and visualizations allow for a meaningful quantitative analysis of the strengths and weaknesses of different algorithms. In contrast to existing evaluation methodologies, application-specific priorities can be taken into account to identify the most suitable algorithms."^^ . "2019" . . . . . . . "Katrin"^^ . "Honauer"^^ . "Katrin Honauer"^^ . . . . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (PDF)"^^ . . . "_dissertation_katrin_honauer.pdf"^^ . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (Other)"^^ . . . . . . "lightbox.jpg"^^ . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (Other)"^^ . . . . . . "preview.jpg"^^ . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (Other)"^^ . . . . . . "medium.jpg"^^ . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (Other)"^^ . . . . . . "small.jpg"^^ . . . "Performance Metrics and Test Data Generation for Depth Estimation Algorithms (Other)"^^ . . . . . . "indexcodes.txt"^^ . . "HTML Summary of #25758 \n\nPerformance Metrics and Test Data Generation for Depth Estimation Algorithms\n\n" . "text/html" . . . "004 Informatik"@de . "004 Data processing Computer science"@en . .