Generic to Specific Recognition Models for Membership Analysis in Group Videos

Abstract

Automatic understanding and analysis of groups has attracted increasing attention in the vision and multimedia communities in recent years. However, little attention has been paid to the automatic analysis of group membership-i.e., recognizing which group the individual in question is part of. This paper presents a novel two-phase Support Vector Machine (SVM) based specific recognition model that is learned using an optimized generic recognition model. We conduct a set of experiments using a database collected to study group analysis from multimodal cues while each group (i.e., four participants together) were watching a number of long movie segments. Our experimental results show that the proposed specific recognition model (52%) outperforms the generic recognition model trained across all different videos (35%) and the independent recognition model trained directly on each specific video (33%) using linear SVM.

A deep generic to specific recognition model for group membership analysis using non-verbal cues

“…In our previous work Mou et al (2017), the generic recognition model and the various specific recognition models were trained separately. Specifically, we first trained a generic recognition model, obtaining an optimal value of the parameter w 0 , and then we trained a set of specific recognition models based on the optimized generic recognition model.…”

“…The video IDs are stated in parentheses and are used to refer to the videos in the rest of the paper; the corresponding conditions and video durations (in minutes) are also listed. tion model, that was proposed in our previous work Mou et al (2017), allows the group membership recognition across all different conditions. However, since group members may behave distinctly in different conditions (e.g., while watching horror movies vs. comedies), the performance of generic recognition model may be significantly limited.…”

“…Therefore, we propose a specific recognition model for each condition specifically, but we learn it on the top of the generic recognition model. This paper is an extended version of our previous work Mou et al (2017). In Mou et al (2017), we proposed a two-phase learning framework to solve the group membership recognition problem, where we first trained a generic recognition model using all videos across all conditions and, then optimized the specific recognition model for each specific condition based on the optimization results obtained from the generic recognition model.…”

“…This paper is an extended version of our previous work Mou et al (2017). In Mou et al (2017), we proposed a two-phase learning framework to solve the group membership recognition problem, where we first trained a generic recognition model using all videos across all conditions and, then optimized the specific recognition model for each specific condition based on the optimization results obtained from the generic recognition model. Different from the aforementioned paper, in this work we unify the generic recognition model and the specific recognition model under a single deep framework.…”

“…Furthermore, we conduct new experiments with a larger dataset. In the rest of the paper, we refer to the specific recognition model (as presented in our previous work Mou et al (2017)) as the two-phase Specific Recognition Model (SRM) and to the proposed Deep Specific Recognition Model as the DeepSRM.…”