ABSTRACT
Temporal action localization (TAL) has drawn much attention in recent years, however, the performance of previous methods is still far from satisfactory due to the lack of annotated untrimmed video data. To deal with this issue, we propose to improve the utilization of current data through feature augmentation. Given an input video, we first extract video features with pre-trained video encoders, and then randomly mask various semantic contents of video features to consider different views of video features. To avoid damaging important action-related semantic information, we further develop a learnable feature augmentation framework to generate better views of videos. In particular, a Mask-based Feature Augmentation Module (MFAM) is proposed. The MFAM has three advantages: 1) it captures the temporal and semantic relationships of original video features, 2) it generates masked features with indispensable action-related information, and 3) it randomly recycles some masked information to ensure diversity. Finally, we input the masked features and the original features into shared action detectors respectively, and perform action classification and localization jointly for model learning. The proposed framework can improve the robustness and generalization of action detectors by learning more and better views of videos. In the testing stage, the MFAM can be removed, which does not bring extra computational costs. Extensive experiments are conducted on four TAL benchmark datasets. Our proposed framework significantly improves different TAL models and achieves the state-of-the-art performances.
ABSTRACT
To solve the problem that the macroscopic deformation and failure of coal-rock medium under external loads are easy to be observed while the internal stress transfer mode and path are unclear. Based on the discrete element idea, the numerical models for pure coal or rock samples and coal-rock combination samples with different lithologies and combination methods under concentrated force are established by PFC2D software. Then the influence of coal or rock strength and combination methods on the internal stress transfer law and distribution evolution characteristics of coal-rock medium are discussed from the perspectives of macroscopic stress and mesoscopic force chain, respectively. The results showed that under concentrated load, the macroscopic stress transfer paths within pure coal or rock samples and coal-rock combination samples are primarily in the form of 'point source radiation'. However, when transferring between coal-rock interfaces, there is a certain interface effect. For pure coal or rock samples, differences in lithology does not change the transfer rules and macro distribution patterns of internal stress, but it can cause changes in internal unit transfer stress value and local area transfer direction. For coal-rock combination samples, the greater the difference in lithology between the two sides of the interface, the more likely the interface effect will occur. In addition, the internal stress transfer is also influenced by the relative stratigraphic relationships of coal and rock. When the stress is transferred from a higher-strength rock to a lower-strength coal mass, the interface effect will be more significant. However, regardless of the combination pattern, the locations where significant stress surges occur are always within the higher strength rock mass near the interface. The findings are helpful to understand the mechanical properties and failure mechanism of mining coal and rock mass, and provide a theoretical basis for the study of the mining-induced mechanical behavior of the floor under the action of the coal pillar.
