Nowadays, the majority of state of the art monocular depth estimation techniques are based on supervised deep learning models. However, collecting RGB images with associated depth maps is a very time consuming procedure. Therefore, recent works have proposed deep architectures for addressing the monocular depth prediction task as a reconstruction problem, thus avoiding the need of collecting ground-truth depth. Following these works, we propose a novel self-supervised deep model for estimating depth maps. Our framework exploits two main strategies: refinement via cycle-inconsistency and distillation. Specifically, first a student network is trained to predict a disparity map such as to recover from a frame in a camera view the associated image in the opposite view. Then, a backward cycle network is applied to the generated image to re-synthesize back the input image, estimating the opposite disparity. A third network exploits the inconsistency between the original and the reconstructed input frame in order to output a refined depth map. Finally, knowledge distillation is exploited, such as to transfer information from the refinement network to the student. Our extensive experimental evaluation demonstrate the effectiveness of the proposed framework which outperforms state of the art unsupervised methods on the KITTI benchmark.

Refine and Distill: Exploiting Cycle-Inconsistency and Knowledge Distillation for Unsupervised Monocular Depth Estimation / Pilzer, Andrea; Lathuiliere, Stephane; Sebe, Nicu; Ricci, Elisa. - (2019), pp. 9760-9769. (Intervento presentato al convegno IEEE Comference on Computer Vision and Pattern Recognition (CVPR'19) tenutosi a Long Beach nel June 16-20, 2019) [10.1109/CVPR.2019.01000].

Refine and Distill: Exploiting Cycle-Inconsistency and Knowledge Distillation for Unsupervised Monocular Depth Estimation

Pilzer, Andrea;Lathuiliere, Stephane;Sebe, Nicu;Ricci, Elisa
2019-01-01

Abstract

Nowadays, the majority of state of the art monocular depth estimation techniques are based on supervised deep learning models. However, collecting RGB images with associated depth maps is a very time consuming procedure. Therefore, recent works have proposed deep architectures for addressing the monocular depth prediction task as a reconstruction problem, thus avoiding the need of collecting ground-truth depth. Following these works, we propose a novel self-supervised deep model for estimating depth maps. Our framework exploits two main strategies: refinement via cycle-inconsistency and distillation. Specifically, first a student network is trained to predict a disparity map such as to recover from a frame in a camera view the associated image in the opposite view. Then, a backward cycle network is applied to the generated image to re-synthesize back the input image, estimating the opposite disparity. A third network exploits the inconsistency between the original and the reconstructed input frame in order to output a refined depth map. Finally, knowledge distillation is exploited, such as to transfer information from the refinement network to the student. Our extensive experimental evaluation demonstrate the effectiveness of the proposed framework which outperforms state of the art unsupervised methods on the KITTI benchmark.
2019
IEEE Comference on Computer Vision and Pattern Recognition (CVPR'19)
New York
IEEE
978-1-7281-3293-8
Pilzer, Andrea; Lathuiliere, Stephane; Sebe, Nicu; Ricci, Elisa
Refine and Distill: Exploiting Cycle-Inconsistency and Knowledge Distillation for Unsupervised Monocular Depth Estimation / Pilzer, Andrea; Lathuiliere, Stephane; Sebe, Nicu; Ricci, Elisa. - (2019), pp. 9760-9769. (Intervento presentato al convegno IEEE Comference on Computer Vision and Pattern Recognition (CVPR'19) tenutosi a Long Beach nel June 16-20, 2019) [10.1109/CVPR.2019.01000].
File in questo prodotto:
File Dimensione Formato  
Pilzer_Refine_and_Distill_Exploiting_Cycle-Inconsistency_and_Knowledge_Distillation_for_Unsupervised_CVPR_2019_paper.pdf

accesso aperto

Tipologia: Post-print referato (Refereed author’s manuscript)
Licenza: Altra licenza (Other type of license)
Dimensione 2.73 MB
Formato Adobe PDF
2.73 MB Adobe PDF Visualizza/Apri
08953744.pdf

Solo gestori archivio

Tipologia: Versione editoriale (Publisher’s layout)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 2.65 MB
Formato Adobe PDF
2.65 MB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11572/250767
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 110
  • ???jsp.display-item.citation.isi??? 75
social impact