1 January 2020 in Information Sciences
1 July 2020 in Information Sciences
28 January 2020 in IEEE Transactions on Cybernetics
Generating action proposals in untrimmed videos is a challenging task, since video sequences usually contain lots of irrelevant contents and the duration of an action instance is arbitrary. The quality of action proposals is key to action detection performance. The previous methods mainly rely on sliding windows or anchor boxes to cover all ground-truth actions, but this is infeasible and computationally inefficient. To this end, this article proposes a RecapNet--a novel framework for generating action proposal, by mimicking the human cognitive process of understanding video content. Specifically, this RecapNet includes a residual causal convolution module to build a short memory of the past events, based on which the joint probability actionness density ranking mechanism is designed to retrieve the action proposals. The RecapNet can handle videos with arbitrary length and more important, a video sequence will need to be processed only in one single pass in order to generate all action proposals. The experiments show that the proposed RecapNet outperforms the state of the art under all metrics on the benchmark THUMOS14 and ActivityNet-1.3 datasets. The code is available publicly at https://github.com/tianwangbuaa/RecapNet.
1 May 2020 in Information Sciences
1 April 2020 in IEEE Access
Applying model-based learning for the optimal decision of the multi-agent system is not trivial due to the expensive price or even the impossibility of obtaining the ground truth for training the model of the complex environment. Such as learning the optimal action of hydraulic supports in the top-coal caving, the optimal action could not accessible as the ground truth of the corresponding state in the intricate processes. Regarding the latent ground truth as the hidden variable is an effective method in the hidden Markov model. This paper extends the hidden variable of ground truth to the multi-agent system and proposes the hidden Markov random field (HMRF) model with reinforcement learning for optimizing the action decision of the multi-agent. In the HMRF model, the input states and the output actions of the multi-agent are considered as an observable random field and a latent Markov random field, respectively. Based on the HMRF model, the optimal decision is inferred by the maximum posterior probability with the prior probability obtained by Q-learning. Meanwhile, the parameters of the HMRF model are estimated by the expectation maximum algorithm. In the experiment, the top-coal caving demonstrates the effectiveness of the proposed method that the recall of top-coal is improved prominently with a very small price of increasing the rock-rate. Furthermore, the proposed method is employed to deal with the predator-preys problem in the gym. The experiment result shows that the communication between agents by the HMRF increases the reward of the preys.