Abstract:The detection of moving infrared dim-small targets has been a challenging and prevalent research topic. The current state-of-the-art methods are mainly based on ConvLSTM to aggregate information from adjacent frames to facilitate the detection of the current frame. However, these methods implicitly utilize motion information only in the training stage and fail to explicitly explore motion compensation, resulting in poor performance in the case of a video sequence including large motion. In this paper, we propose a Deformable Feature Alignment and Refinement (DFAR) method based on deformable convolution to explicitly use motion context in both the training and inference stages. Specifically, a Temporal Deformable Alignment (TDA) module based on the designed Dilated Convolution Attention Fusion (DCAF) block is developed to explicitly align the adjacent frames with the current frame at the feature level. Then, the feature refinement module adaptively fuses the aligned features and further aggregates useful spatio-temporal information by means of the proposed Attention-guided Deformable Fusion (AGDF) block. In addition, to improve the alignment of adjacent frames with the current frame, we extend the traditional loss function by introducing a new motion compensation loss. Extensive experimental results demonstrate that the proposed DFAR method achieves the state-of-the-art performance on two benchmark datasets including DAUB and IRDST.
Abstract:Recent interests in dynamic decision modeling have led to the development of several representation and inference methods. These methods however, have limited application under time critical conditions where a trade-off between model quality and computational tractability is essential. This paper presents an approach to time-critical dynamic decision modeling. A knowledge representation and modeling method called the time-critical dynamic influence diagram is proposed. The formalism has two forms. The condensed form is used for modeling and model abstraction, while the deployed form which can be converted from the condensed form is used for inference purposes. The proposed approach has the ability to represent space-temporal abstraction within the model. A knowledge-based meta-reasoning approach is proposed for the purpose of selecting the best abstracted model that provide the optimal trade-off between model quality and model tractability. An outline of the knowledge-based model construction algorithm is also provided.
Abstract:As Bayesian networks are applied to larger and more complex problem domains, search for flexible modeling and more efficient inference methods is an ongoing effort. Multiply sectioned Bayesian networks (MSBNs) extend the HUGIN inference for Bayesian networks into a coherent framework for flexible modeling and distributed inference.Lazy propagation extends the Shafer-Shenoy and HUGIN inference methods with reduced space complexity. We apply the Shafer-Shenoy and lazy propagation to inference in MSBNs. The combination of the MSBN framework and lazy propagation provides a better framework for modeling and inference in very large domains. It retains the modeling flexibility of MSBNs and reduces the runtime space complexity, allowing exact inference in much larger domains given the same computational resources.