Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation
Depth perception capability is one of the essential requirements for various autonomous driving platforms. However, accurate depth estimation in a real-world setting is still a challenging problem due to high computational costs. In this paper, we propose a lightweight depth completion network for d...
Main Authors: | , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2022-09-01
|
Series: | Sensors |
Subjects: | |
Online Access: | https://www.mdpi.com/1424-8220/22/19/7388 |
_version_ | 1797476897232257024 |
---|---|
author | Yongseop Jeong Jinsun Park Donghyeon Cho Yoonjin Hwang Seibum B. Choi In So Kweon |
author_facet | Yongseop Jeong Jinsun Park Donghyeon Cho Yoonjin Hwang Seibum B. Choi In So Kweon |
author_sort | Yongseop Jeong |
collection | DOAJ |
description | Depth perception capability is one of the essential requirements for various autonomous driving platforms. However, accurate depth estimation in a real-world setting is still a challenging problem due to high computational costs. In this paper, we propose a lightweight depth completion network for depth perception in real-world environments. To effectively transfer a teacher’s knowledge, useful for the depth completion, we introduce local similarity-preserving knowledge distillation (LSPKD), which allows similarities between local neighbors to be transferred during the distillation. With our LSPKD, a lightweight student network is precisely guided by a heavy teacher network, regardless of the density of the ground-truth data. Experimental results demonstrate that our method is effective to reduce computational costs during both training and inference stages while achieving superior performance over other lightweight networks. |
first_indexed | 2024-03-09T21:11:15Z |
format | Article |
id | doaj.art-916375e056364ba3acc3772ad43d4550 |
institution | Directory Open Access Journal |
issn | 1424-8220 |
language | English |
last_indexed | 2024-03-09T21:11:15Z |
publishDate | 2022-09-01 |
publisher | MDPI AG |
record_format | Article |
series | Sensors |
spelling | doaj.art-916375e056364ba3acc3772ad43d45502023-11-23T21:48:22ZengMDPI AGSensors1424-82202022-09-012219738810.3390/s22197388Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge DistillationYongseop Jeong0Jinsun Park1Donghyeon Cho2Yoonjin Hwang3Seibum B. Choi4In So Kweon5The Robotics Program, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, KoreaSchool of Computer Science and Engineering, Pusan National University, 2 Busandaehak-ro 63beon-gil, Geumjeong-gu, Busan 46241, KoreaDepartment of Electronics Engineering, Chungnam National University, 99 Daehak-ro, Yuseong-gu, Daejeon 34134, KoreaDepartment of Mechanical Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, KoreaDepartment of Mechanical Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, KoreaSchool of Electrical Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, KoreaDepth perception capability is one of the essential requirements for various autonomous driving platforms. However, accurate depth estimation in a real-world setting is still a challenging problem due to high computational costs. In this paper, we propose a lightweight depth completion network for depth perception in real-world environments. To effectively transfer a teacher’s knowledge, useful for the depth completion, we introduce local similarity-preserving knowledge distillation (LSPKD), which allows similarities between local neighbors to be transferred during the distillation. With our LSPKD, a lightweight student network is precisely guided by a heavy teacher network, regardless of the density of the ground-truth data. Experimental results demonstrate that our method is effective to reduce computational costs during both training and inference stages while achieving superior performance over other lightweight networks.https://www.mdpi.com/1424-8220/22/19/7388depth completionlocal similarityknowledge distillationmodel compressionsensor fusionmultimodal learning |
spellingShingle | Yongseop Jeong Jinsun Park Donghyeon Cho Yoonjin Hwang Seibum B. Choi In So Kweon Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation Sensors depth completion local similarity knowledge distillation model compression sensor fusion multimodal learning |
title | Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation |
title_full | Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation |
title_fullStr | Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation |
title_full_unstemmed | Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation |
title_short | Lightweight Depth Completion Network with Local Similarity-Preserving Knowledge Distillation |
title_sort | lightweight depth completion network with local similarity preserving knowledge distillation |
topic | depth completion local similarity knowledge distillation model compression sensor fusion multimodal learning |
url | https://www.mdpi.com/1424-8220/22/19/7388 |
work_keys_str_mv | AT yongseopjeong lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation AT jinsunpark lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation AT donghyeoncho lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation AT yoonjinhwang lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation AT seibumbchoi lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation AT insokweon lightweightdepthcompletionnetworkwithlocalsimilaritypreservingknowledgedistillation |