Unsupervised Domain Adaptation for Depth Prediction from Images

State-of-the-art approaches to infer dense depth measurements from images rely on CNNs trained end-to-end on a vast amount of data. However, these approaches suffer a drastic drop in accuracy when dealing with environments much different in appearance and/or context from those observed at training t...

Ausführliche Beschreibung

Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on pattern analysis and machine intelligence. - 1979. - 42(2020), 10 vom: 06. Okt., Seite 2396-2409
1. Verfasser: Tonioni, Alessio (VerfasserIn)
Weitere Verfasser: Poggi, Matteo, Mattoccia, Stefano, Stefano, Luigi Di
Format: Online-Aufsatz
Sprache:English
Veröffentlicht: 2020
Zugriff auf das übergeordnete Werk:IEEE transactions on pattern analysis and machine intelligence
Schlagworte:Journal Article Research Support, Non-U.S. Gov't
LEADER 01000naa a22002652 4500
001 NLM301226881
003 DE-627
005 20231225104207.0
007 cr uuu---uuuuu
008 231225s2020 xx |||||o 00| ||eng c
024 7 |a 10.1109/TPAMI.2019.2940948  |2 doi 
028 5 2 |a pubmed24n1004.xml 
035 |a (DE-627)NLM301226881 
035 |a (NLM)31514127 
040 |a DE-627  |b ger  |c DE-627  |e rakwb 
041 |a eng 
100 1 |a Tonioni, Alessio  |e verfasserin  |4 aut 
245 1 0 |a Unsupervised Domain Adaptation for Depth Prediction from Images 
264 1 |c 2020 
336 |a Text  |b txt  |2 rdacontent 
337 |a ƒaComputermedien  |b c  |2 rdamedia 
338 |a ƒa Online-Ressource  |b cr  |2 rdacarrier 
500 |a Date Completed 12.02.2021 
500 |a Date Revised 12.02.2021 
500 |a published: Print-Electronic 
500 |a Citation Status PubMed-not-MEDLINE 
520 |a State-of-the-art approaches to infer dense depth measurements from images rely on CNNs trained end-to-end on a vast amount of data. However, these approaches suffer a drastic drop in accuracy when dealing with environments much different in appearance and/or context from those observed at training time. This domain shift issue is usually addressed by fine-tuning on smaller sets of images from the target domain annotated with depth labels. Unfortunately, relying on such supervised labeling is seldom feasible in most practical settings. Therefore, we propose an unsupervised domain adaptation technique which does not require groundtruth labels. Our method relies only on image pairs and leverages on classical stereo algorithms to produce disparity measurements alongside with confidence estimators to assess upon their reliability. We propose to fine-tune both depth-from-stereo as well as depth-from-mono architectures by a novel confidence-guided loss function that handles the measured disparities as noisy labels weighted according to the estimated confidence. Extensive experimental results based on standard datasets and evaluation protocols prove that our technique can address effectively the domain shift issue with both stereo and monocular depth prediction architectures and outperforms other state-of-the-art unsupervised loss functions that may be alternatively deployed to pursue domain adaptation 
650 4 |a Journal Article 
650 4 |a Research Support, Non-U.S. Gov't 
700 1 |a Poggi, Matteo  |e verfasserin  |4 aut 
700 1 |a Mattoccia, Stefano  |e verfasserin  |4 aut 
700 1 |a Stefano, Luigi Di  |e verfasserin  |4 aut 
773 0 8 |i Enthalten in  |t IEEE transactions on pattern analysis and machine intelligence  |d 1979  |g 42(2020), 10 vom: 06. Okt., Seite 2396-2409  |w (DE-627)NLM098212257  |x 1939-3539  |7 nnns 
773 1 8 |g volume:42  |g year:2020  |g number:10  |g day:06  |g month:10  |g pages:2396-2409 
856 4 0 |u http://dx.doi.org/10.1109/TPAMI.2019.2940948  |3 Volltext 
912 |a GBV_USEFLAG_A 
912 |a SYSFLAG_A 
912 |a GBV_NLM 
912 |a GBV_ILN_350 
951 |a AR 
952 |d 42  |j 2020  |e 10  |b 06  |c 10  |h 2396-2409