Monitored Distillation for Positive Congruent Depth Completion
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Liu, T.Y. | - |
dc.contributor.author | Agrawal, P. | - |
dc.contributor.author | Chen, A. | - |
dc.contributor.author | Hong, Byung-Woo | - |
dc.contributor.author | Wong, A. | - |
dc.date.accessioned | 2022-12-16T06:41:54Z | - |
dc.date.available | 2022-12-16T06:41:54Z | - |
dc.date.issued | 2022-10 | - |
dc.identifier.issn | 0302-9743 | - |
dc.identifier.issn | 1611-3349 | - |
dc.identifier.uri | https://scholarworks.bwise.kr/cau/handle/2019.sw.cau/59679 | - |
dc.description.abstract | We propose a method to infer a dense depth map from a single image, its calibration, and the associated sparse point cloud. In order to leverage existing models (teachers) that produce putative depth maps, we propose an adaptive knowledge distillation approach that yields a positive congruent training process, wherein a student model avoids learning the error modes of the teachers. In the absence of ground truth for model selection and training, our method, termed Monitored Distillation, allows a student to exploit a blind ensemble of teachers by selectively learning from predictions that best minimize the reconstruction error for a given image. Monitored Distillation yields a distilled depth map and a confidence map, or “monitor”, for how well a prediction from a particular teacher fits the observed image. The monitor adaptively weights the distilled depth where if all of the teachers exhibit high residuals, the standard unsupervised image reconstruction loss takes over as the supervisory signal. On indoor scenes (VOID), we outperform blind ensembling baselines by 17.53% and unsupervised methods by 24.25%; we boast a 79% model size reduction while maintaining comparable performance to the best supervised method. For outdoors (KITTI), we tie for 5th overall on the benchmark despite not using ground truth. Code available at: https://github.com/alexklwong/mondi-python. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG. | - |
dc.format.extent | 19 | - |
dc.language | 영어 | - |
dc.language.iso | ENG | - |
dc.publisher | Springer Science and Business Media Deutschland GmbH | - |
dc.title | Monitored Distillation for Positive Congruent Depth Completion | - |
dc.type | Article | - |
dc.identifier.doi | 10.1007/978-3-031-20086-1_3 | - |
dc.identifier.bibliographicCitation | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), v.13662 LNCS, pp 35 - 53 | - |
dc.description.isOpenAccess | N | - |
dc.identifier.wosid | 000899248700003 | - |
dc.identifier.scopusid | 2-s2.0-85142749457 | - |
dc.citation.endPage | 53 | - |
dc.citation.startPage | 35 | - |
dc.citation.title | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | - |
dc.citation.volume | 13662 LNCS | - |
dc.type.docType | Proceedings Paper | - |
dc.publisher.location | 미국 | - |
dc.subject.keywordAuthor | Blind ensemble | - |
dc.subject.keywordAuthor | Depth completion | - |
dc.subject.keywordAuthor | Knowledge distillation | - |
dc.relation.journalResearchArea | Computer Science | - |
dc.relation.journalResearchArea | Imaging Science & Photographic Technology | - |
dc.relation.journalWebOfScienceCategory | Computer Science, Artificial Intelligence | - |
dc.relation.journalWebOfScienceCategory | Imaging Science & Photographic Technology | - |
dc.description.journalRegisteredClass | scopus | - |
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.
84, Heukseok-ro, Dongjak-gu, Seoul, Republic of Korea (06974)02-820-6194
COPYRIGHT 2019 Chung-Ang University All Rights Reserved.
Certain data included herein are derived from the © Web of Science of Clarivate Analytics. All rights reserved.
You may not copy or re-distribute this material in whole or in part without the prior written consent of Clarivate Analytics.