The Experts below are selected from a list of 282 Experts worldwide ranked by ideXlab platform
Radoslav Skoviera - One of the best experts on this subject based on the ideXlab platform.
-
exploring Logical Consistency and viewport sensitivity in compositional vqa models
Intelligent Robots and Systems, 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
-
IROS - Exploring Logical Consistency and viewport sensitivity in compositional VQA models
2019 IEEE RSJ International Conference on Intelligent Robots and Systems (IROS), 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
Gabriela Sejnova - One of the best experts on this subject based on the ideXlab platform.
-
exploring Logical Consistency and viewport sensitivity in compositional vqa models
Intelligent Robots and Systems, 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
-
IROS - Exploring Logical Consistency and viewport sensitivity in compositional VQA models
2019 IEEE RSJ International Conference on Intelligent Robots and Systems (IROS), 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
Benjamin M Craig - One of the best experts on this subject based on the ideXlab platform.
-
Relative risk of a shuffled deck: a generalizable Logical Consistency criterion for sample selection in health state valuation studies.
Health economics, 2006Co-Authors: Benjamin M Craig, Sulabha RamachandranAbstract:In a health state valuation study, respondents may be asked to rank a deck of cards, with each card representing a particular health state. A Logical inConsistency occurs when a more severe health state card is ranked higher than a less severe card. Occasional inconsistencies may be justified by errors in judgment or measurement. However, when respondents return shuffled decks, their responses must be removed from the sample; otherwise, valuation estimates will be biased toward the median. In this paper, we present a Logical Consistency criterion for sample selection in health state valuation studies. This statistical criterion is based on the relative risk of a shuffled deck and generalizable to all health state classification systems, subsets (or decks) of health states, and valuation techniques. We applied the criterion to secondary data collected from 4048 United States and 3395 United Kingdom respondents. In both studies, respondents evaluated 12-card decks of EQ-5D health states using time trade-off and visual analog scale techniques. Among the UK respondents, a small portion (approximately 5%) did not satisfy the criterion; their exclusion significantly changed the sample characteristics and the mean value estimates of the EQ-5D health states. Similar results were found among the US respondents.
-
ql2 Logical Consistency and the valuation of health an analysis of us survey data
Value in Health, 2001Co-Authors: Benjamin M Craig, Paul KindAbstract:OBJECTIVES: It is widely held that values of the general public should be used in the evaluation of health care. Surveys designed to record such values involve the participation of individuals with different health experiences and with different socioeconomic backgrounds. The technical performance of these participants is likely to vary as a function of these factors, for example the Logical Consistency of responses is often associated with socioeconomic status. This paper examines the relationship between Logical Consistency and respondent health using US survey data designed to capture values for states defined by the EQ-5D classification. METHODS: A standardised questionnaire was used to elicit valuations for EQ-5D health states in a postal survey conducted by Johnson et al (1998, Pharmacoeconomics) in Arizona in which US respondents (N = 905) rated eight states along a visual analog scale from best to worst imaginable health. A Logical ordering is defined for 23 unique pairs of states in that one state dominates the other over all 5 dimension of the EQ-5D. A Logical inConsistency was noted when a respondent assigned a lower value to the “better” state in such a pair. Censored regression models were used to assess the relationship between Consistency and respondent health. We tested the robustness of these findings using survey data from Wisconsin, which applied the same questionnaire (N = 222). RESULTS: From the best imaginable state, each 20-point decrement in respondent's self-rated health status yielded significantly greater inConsistency in their valuation of EQ-5D health states controlling for age and sex. Inclusion of education and income reduced this effect slightly, yet it remained statistically significant. CONCLUSIONS: Respondents in poor health demonstrate greater difficult in valuing health states in a Logically consistent manner. Censoring survey data to remove inconsistent respondents may violate the principle of using representative population values in evaluating cost-effectiveness of health care.
Michal Vavrecka - One of the best experts on this subject based on the ideXlab platform.
-
exploring Logical Consistency and viewport sensitivity in compositional vqa models
Intelligent Robots and Systems, 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
-
IROS - Exploring Logical Consistency and viewport sensitivity in compositional VQA models
2019 IEEE RSJ International Conference on Intelligent Robots and Systems (IROS), 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
Michael Tesar - One of the best experts on this subject based on the ideXlab platform.
-
exploring Logical Consistency and viewport sensitivity in compositional vqa models
Intelligent Robots and Systems, 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
-
IROS - Exploring Logical Consistency and viewport sensitivity in compositional VQA models
2019 IEEE RSJ International Conference on Intelligent Robots and Systems (IROS), 2019Co-Authors: Gabriela Sejnova, Michal Vavrecka, Michael Tesar, Radoslav SkovieraAbstract:The most recent architectures for Visual Question Answering (VQA), such as TbD or DDRprog, have already outperformed human-level accuracy on benchmark datasets (e.g. CLEVR). We administered an advanced analysis of their performance based on novel metrics called Consistency (sum of all object feature instances in the scene (e.g. shapes) equals the total number of the objects in the scene) and revealed only 56% Consistency for the most accurate architecture (TbD). In respect to this finding, we propose a new method of the VQA training, which reaches 98% Consistency. Furthermore, testing of the VQA model in real world brings out a problem with precise mimicking of the camera position from the original dataset. We therefore created a virtual environment along with its real-world counterpart with variable camera positions to test the accuracy and Consistency from different viewports. Based on these errors, we were able to estimate optimal position of the camera. The proposed method thus allows us to find the optimal camera viewport in the real environment without knowing the geometry and the exact position of the camera in the synthetic training environment.
