In this paper, a multi-object recognition scenario is considered to extend the random finite set into random permutation set. Probabilistic information on random permutation set can be viewed as an distribution determined by three random variables. We use another emerging uncertainty representation, order-2 information granule, to realize the probabilistic information fusion on random permutation sets. First, the probabilistic information on random permutation sets is viewed as an order-2 probability distribution. Second, corresponding information fusion approach is proposed. Finally, the proposed approach is applied to random permutation sets, resolving the decision-making issue under the multi-object recognition scenario. This paper pioneers the connection of order-2 information processing logic to a multi-object recognition task and develops order-2 probability distribution and its combination rules. Compared to the traditional probabilistic information fusion approaches, the proposed approach takes into account not only the propositions’ beliefs provided by the sources, but the structural dependency among propositions as well.

Based on subjective possibilistic semantics, an agent's subjective probability mass function is dominated by a qualitative Possibility Mass Function (PossMF), which can also be transformed into a unique consonant mass function. However, the existing transformation method cannot maintain the consistency of combination rules, i.e., fusing PossMFs and consonant mass functions with same information content, respectively, the results no longer maintain the reversible transformation. To address the above issue, a novel belief functions transformation is proposed, which can be interpreted based on both Smets' canonical decomposition and Pichon's canonical decomposition. The proposed method is validated based on consistency of combination rules, the least commitment principle, and its application in the fusion of information. In addition, based on the two canonical decompositions, we extend the transformation to possibilistic belief structure, and offer a new perspective of relationship between possibilistic information and evidential information.

Natural language tracking aims to localize the target object referred to by a language description using a sequence of bounding boxes in video frames. Compared with traditional visual single object tracking initialized with only a bounding box (BBox), this task introduces high-level semantic information to reduce the ambiguity of BBox and enhance the ability to retrieve the target in a global manner. Thus, it can yield more accurate and robust tracking results. Previous methods usually adopt off-the-shelf grounding and tracking branches to tackle this task, where feature representations are learned in isolation without benefiting each other. The two branches can associate with each other to discover crucial clues since the language description and the template image provide information from different sources. Therefore, we propose a unified transformer method for natural language tracking named TransNLT, which utilizes isomorphic Transformer structures for grounding and tracking branches where collaborative learning is enabled to construct comprehensive features for the target. In addition, we propose a Selective Feature Gathering (SFG) module which can integrate cross-modal global information of the tracking target from the visual template and language description. Through effective interaction of visual and language information, we can achieve better results than tracking with only a single modality. Extensive experiments on three popular natural language tracking benchmarks show our proposed TransNLT outperforms previous state-of-the-art methods.

Quantifying the uncertainty of information distributions containing randomness, imprecision, and fuzziness is the premise of processing them. A useful information representation in the field of intelligent computing are information granules, which optimize data from the perspective of specificity and coverage. We introduce information granularity into evidential information and model the basic probability assignment (BPA) as a weighted information granules model. Based on the proposed model, a new uncertainty measure of BPA is derived from the quality evaluation of granules. In addition, the proposed measure is extended to fuzzy evidential information distributions. When the Fuzzy BPA (FBPA) degenerates into the Probability Mass Function (ProbMF) and Possibility Mass Function (PossMF), the proposed method degenerates to Gini entropy and Yager's specificity measure, respectively. We use a refined belief structure to interpret the meaning of FBPA in the transfer belief model, and verify the validity of the proposed method by analyzing its properties and presenting numerical examples. The concept of information granule is used for the first time to model focal set and beliefs. Compared with Shannon entropy based information measures, the proposed method provides a novel perspective on the relationship between randomness, imprecision, and fuzziness in FBPA.

Based on Klir’s framework of uncertainty, the total uncertainty (also called ambiguity) of belief function is linear addition of discord and nonspecificity. Though uncertainty measures of belief function have been discussed widely, there is no measure that can satisfy the monotonicity and range consistency properties at the same time. In this article, we discuss uncertainty measure of belief function from the perspective of information fractal dimension. An uncertainty quantity called evenness and its measure Eve are proposed, which can represent the belief propensity degree of belief function. We first propose the measures of diversity (normalized nonspecificity) and the element evenness (normalized discord), and then fuse them to calculate Eve. The proposed method can not only measure the subnormal mass function but also interpret the different views of Klir and Smets on “Uncertainty.” In addition, we extend Klir’s framework of uncertainty based on the proposed information quantities.

The goal of referring image segmentation is to identify the object matched with an input natural language expression. Previous methods only support English descriptions, whereas Chinese is also broadly used around the world, which limits the potential application of this task. Therefore, we propose to extend existing datasets with Chinese descriptions and preprocessing tools for training and evaluating bilingual referring segmentation models. In addition, previous methods also lack the ability to collaboratively learn channel-wise and spatial-wise cross-modal attention to well align visual and linguistic modalities. To tackle these limitations, we propose a Linguistic Excitation module to excite image channels guided by language information and a Linguistic Aggregation module to aggregate multimodal information based on image-language relationships. Since different levels of features from the visual backbone encode rich visual information, we also propose a Cross-Level Attentive Fusion module to fuse multilevel features gated by language information. Extensive experiments on four English and Chinese benchmarks show that our bilingual referring image segmentation model outperforms previous methods.

Pedestrian attribute recognition, which can benefit other tasks such as person re-identification and pedestrian retrieval, is very important in video surveillance related tasks. In this paper, we observe that the existing methods tackle this problem from the perspective of multi-label classification without considering the hierarchical relationships among the attributes. In human cognition, the attributes can be categorized according to their semantic/abstraction levels. The high-level attributes can be predicted by reasoning from the low-level and medium-level attributes, while the recognition of the low-level and medium-level attributes can be guided by the high-level attributes. Based on this attribute categorization, we propose a novel Hierarchical Reasoning Network (HR-Net), which can hierarchically predict the attributes at different abstraction levels in different stages of the network. We also propose an attribute reasoning structure to exploit the relationships among the attributes at different semantic levels. Experimental results demonstrate that the proposed network gives superior performances compared to the state-of-the-art techniques.