Temporal Model Adaptation for Person Re-identification
Person re-identification is an open and challenging problem in computer vision. Majority of the efforts have been spent either to design the best feature representation or to learn the optimal matching metric. Most approaches have neglected the problem of
- PDF / 1,184,244 Bytes
- 20 Pages / 439.37 x 666.142 pts Page_size
- 18 Downloads / 189 Views
2 3
University of Udine, 33100 Udine, Italy [email protected] University of Massatchussets Lowell, Lowell, MA 01852, USA University of California Riverside, Riverside, CA 92507, USA
Abstract. Person re-identification is an open and challenging problem in computer vision. Majority of the efforts have been spent either to design the best feature representation or to learn the optimal matching metric. Most approaches have neglected the problem of adapting the selected features or the learned model over time. To address such a problem, we propose a temporal model adaptation scheme with human in the loop. We first introduce a similarity-dissimilarity learning method which can be trained in an incremental fashion by means of a stochastic alternating directions methods of multipliers optimization procedure. Then, to achieve temporal adaptation with limited human effort, we exploit a graph-based approach to present the user only the most informative probe-gallery matches that should be used to update the model. Results on three datasets have shown that our approach performs on par or even better than state-of-the-art approaches while reducing the manual pairwise labeling effort by about 80 %.
Keywords: Person re-identificaion
1
· Metric learning · Active learning
Introduction
Person re-identification is the problem of matching a person acquired by disjoint cameras at different time instants. The problem has recently gained increasing attention (see [1] for a recent survey) due to its open challenges like changes in viewing angle, background clutter, and occlusions. To address these issues, existing approaches seek either the best feature representations (e.g., [2–4]) or propose to learn optimal matching metrics (e.g., [5–7]). While they have obtained reasonable performance on commonly used datasets (e.g., [8–10]), we believe that these approaches have not yet considered a fundamental related problem: how to learn from the data being continuously collected in an installed system and Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46493-0 52) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part IV, LNCS 9908, pp. 858–877, 2016. DOI: 10.1007/978-3-319-46493-0 52
Temporal Model Adaptation for Person Re-identification Training Images
( , )( , ) ( , )( , )
OUR CONTRIBUTION Model Training
Incremental Update
Reduced Effort Human Labeling
?
Training Pairs
Deployment Images
( (
, ,
859
) )
Model
( (
, ,
?
) )
Probe-Gallery Pairs for Match
Fig. 1. Illustration of the re-identification pipeline highlighting our contribution. Dashed lines indicate the training stage, solid lines the deployment stage. Existing methods do not consider the information provided by a matched probe-gallery pair to update the model. We propose to use such information to improve the model performance by adapting it to the dynamic environmental variations.
adapt existing models to this new
Data Loading...
