Weakly supervised temporal action localization with additional sub networks for local spatial information
Abstract: In this paper, we propose a weakly supervised learning model for temporal action localization using spatial information. Our proposed model extends STPN by adding sub networks inspired by the patch classification branches. While STPN uses only feature vectors based on 3D CNN, the proposed model also uses frame-level features to reduce the false positives of actions. The effectiveness of the proposed model is demonstrated by experimental results with quantitative and qualitative comparisons to the STPN model.
0 Replies
Loading