FN Archimer Export Format
PT J
TI Homography-based loss function for camera pose regression
BT 
AF Boittiaux, Clementin
   Marxer, Ricard
   Dune, Claire
   Arnaubec, Aurelien
   Hugel, Vincent
AS 1:1,2,3;2:2;3:3;4:1;5:3;
FF 1:PDG-DFO-SM-PRAO;2:;3:;4:PDG-DFO-SM-PRAO;5:;
C1 Ifremer, 83000 La Seyne-sur-Mer, France
   CNRS, LIS, Université de Toulon, Aix Marseille Université, 83000 Toulon, France
   COSMER, Université de Toulon, 83000 Toulon, France
C2 IFREMER, FRANCE
   CNRS, FRANCE
   UNIV TOULON, FRANCE
SI TOULON
SE PDG-DFO-SM-PRAO
IN WOS Ifremer UPR
   copubli-france
   copubli-univ-france
IF 5.2
TC 3
UR https://archimer.ifremer.fr/doc/00766/87810/93679.pdf
LA English
DT Article
DE ;Localization;deep learning for visual perception
AB Some recent visual-based relocalization algorithms rely on deep-learning methods to perform camera pose regression from image data. This paper focuses on the loss functions that embed the error between two poses to perform deep-learning based camera pose regression. Existing loss functions are either difficult-to-tune multi-objective functions or present unstable reprojection errors that rely on ground-truth 3D scene points and require a two-step training. To deal with these issues, we introduce a novel loss function which is based on a multiplane homography integration. This new function does not require prior initialization and only depends on physically interpretable hyperparameters. Furthermore, the experiments carried out on well established relocalization datasets show that it minimizes best the mean square reprojection error during training when compared with existing loss functions. 
PY 2022
PD JUN
SO Ieee Robotics And Automation Letters
SN 2377-3766
PU Institute of Electrical and Electronics Engineers (IEEE)
VL 7
IS 3
UT 000788948000002
BP 6242
EP 6249
DI 10.1109/LRA.2022.3168329
ID 87810
ER

EF