Matches in SemOpenAlex for { <https://semopenalex.org/work/W4320921420> ?p ?o ?g. }
- W4320921420 endingPage "110389" @default.
- W4320921420 startingPage "110389" @default.
- W4320921420 abstract "Supervised data are not always uncontroversial, especially for boundary detection tasks. Considering that we have a portrait, the face contour is naturally the most salient boundary. Interestingly, some people, but not everyone, regard their eyes or eyebrows as boundaries. This means that humans might have different understandings of the same thing, which leads to nondeterministic labels. In this paper, we propose a novel head function based on the Beta distribution for boundary detection. Different from learning the probability in the Bernoulli distribution, it introduces more abundant information. It can be viewed as the distribution of the former’s parameters and captures the uncertainty for binary classification problems. To this end, we employ the maximum likelihood and knowledge distillation as loss functions to train a deep network with the Beta head function. Moreover, we propose an efficient data processing method based on a recurrent voting strategy for merging these nondeterministic labels. In implementation, our Beta head function is lightweight and not limited to specific models. It can be seamlessly employed in many existing boundary detection networks without modifying their backbones. After introducing our Beta head function, the performances of three well-known boundary detection networks have been obviously improved. Multiple experiments indicate the effectiveness and the advantage of introducing uncertainty by the proposed Beta head function." @default.
- W4320921420 created "2023-02-16" @default.
- W4320921420 creator A5017517029 @default.
- W4320921420 creator A5061505268 @default.
- W4320921420 creator A5087615637 @default.
- W4320921420 date "2023-04-01" @default.
- W4320921420 modified "2023-09-23" @default.
- W4320921420 title "Beta network for boundary detection under nondeterministic labels" @default.
- W4320921420 cites W105270443 @default.
- W4320921420 cites W125693051 @default.
- W4320921420 cites W1539790486 @default.
- W4320921420 cites W1910619957 @default.
- W4320921420 cites W1929880813 @default.
- W4320921420 cites W1930528368 @default.
- W4320921420 cites W1946445415 @default.
- W4320921420 cites W1976047850 @default.
- W4320921420 cites W2110158442 @default.
- W4320921420 cites W2111959010 @default.
- W4320921420 cites W2119823327 @default.
- W4320921420 cites W2145023731 @default.
- W4320921420 cites W2151049637 @default.
- W4320921420 cites W2194775991 @default.
- W4320921420 cites W2257235457 @default.
- W4320921420 cites W2289871033 @default.
- W4320921420 cites W2300687442 @default.
- W4320921420 cites W2330485005 @default.
- W4320921420 cites W2483076098 @default.
- W4320921420 cites W2553156677 @default.
- W4320921420 cites W2560622558 @default.
- W4320921420 cites W2739879705 @default.
- W4320921420 cites W2783538964 @default.
- W4320921420 cites W2953881420 @default.
- W4320921420 cites W2954996726 @default.
- W4320921420 cites W2963946669 @default.
- W4320921420 cites W2979455868 @default.
- W4320921420 cites W2980246484 @default.
- W4320921420 cites W2990284741 @default.
- W4320921420 cites W3009158303 @default.
- W4320921420 cites W3010815318 @default.
- W4320921420 cites W3011680720 @default.
- W4320921420 cites W3034368386 @default.
- W4320921420 cites W3093425017 @default.
- W4320921420 cites W3102100346 @default.
- W4320921420 cites W3118207393 @default.
- W4320921420 cites W3146366485 @default.
- W4320921420 cites W3176913189 @default.
- W4320921420 cites W3187359482 @default.
- W4320921420 cites W3192537953 @default.
- W4320921420 cites W3216361847 @default.
- W4320921420 cites W4205319111 @default.
- W4320921420 cites W4214774769 @default.
- W4320921420 cites W4225257078 @default.
- W4320921420 cites W4283781662 @default.
- W4320921420 cites W4288069699 @default.
- W4320921420 cites W4312710812 @default.
- W4320921420 cites W845365781 @default.
- W4320921420 doi "https://doi.org/10.1016/j.knosys.2023.110389" @default.
- W4320921420 hasPublicationYear "2023" @default.
- W4320921420 type Work @default.
- W4320921420 citedByCount "0" @default.
- W4320921420 crossrefType "journal-article" @default.
- W4320921420 hasAuthorship W4320921420A5017517029 @default.
- W4320921420 hasAuthorship W4320921420A5061505268 @default.
- W4320921420 hasAuthorship W4320921420A5087615637 @default.
- W4320921420 hasConcept C105795698 @default.
- W4320921420 hasConcept C11413529 @default.
- W4320921420 hasConcept C114793014 @default.
- W4320921420 hasConcept C119857082 @default.
- W4320921420 hasConcept C12267149 @default.
- W4320921420 hasConcept C127313418 @default.
- W4320921420 hasConcept C127413603 @default.
- W4320921420 hasConcept C134306372 @default.
- W4320921420 hasConcept C14036430 @default.
- W4320921420 hasConcept C146978453 @default.
- W4320921420 hasConcept C152361515 @default.
- W4320921420 hasConcept C153180895 @default.
- W4320921420 hasConcept C154945302 @default.
- W4320921420 hasConcept C176181172 @default.
- W4320921420 hasConcept C21621910 @default.
- W4320921420 hasConcept C2780312720 @default.
- W4320921420 hasConcept C33923547 @default.
- W4320921420 hasConcept C41008148 @default.
- W4320921420 hasConcept C62354387 @default.
- W4320921420 hasConcept C66905080 @default.
- W4320921420 hasConcept C78458016 @default.
- W4320921420 hasConcept C86803240 @default.
- W4320921420 hasConceptScore W4320921420C105795698 @default.
- W4320921420 hasConceptScore W4320921420C11413529 @default.
- W4320921420 hasConceptScore W4320921420C114793014 @default.
- W4320921420 hasConceptScore W4320921420C119857082 @default.
- W4320921420 hasConceptScore W4320921420C12267149 @default.
- W4320921420 hasConceptScore W4320921420C127313418 @default.
- W4320921420 hasConceptScore W4320921420C127413603 @default.
- W4320921420 hasConceptScore W4320921420C134306372 @default.
- W4320921420 hasConceptScore W4320921420C14036430 @default.
- W4320921420 hasConceptScore W4320921420C146978453 @default.
- W4320921420 hasConceptScore W4320921420C152361515 @default.
- W4320921420 hasConceptScore W4320921420C153180895 @default.