Recently, advances in person re-identification (Re-ID) has benefitted from use of the popular multi-branch network. However, performing feature learning in a single branch with uniform partitioning is likely to separate meaningful local regions, and correlation among different branches is not well established. In this article, we propose a novel harmonious multi-branch network (HMBN) to relieve these intra-branch and inter-branch problems harmoniously. HMBN is a multi-branch network with various stripes on different branches to learn coarse-to-fine pedestrian information. We first replace the uniform partition with a horizontal overlapped partition to cover meaningful local regions between adjacent stripes in a single branch. We then incorporate a novel attention module to make all branches interact by modeling spatial contextual dependencies across branches. Finally, in order to train the HMBN more effectively, a harder triplet loss is introduced to optimize triplets in a harder manner. Extensive experiments are conducted on three benchmark datasets — DukeMTMC-reID, CUHK03, and Market-1501 — demonstrating the superiority of our proposed HMBN over state-of-the-art methods.

Introducing knowledge graphs as side information into recommendation can effectively mitigate data sparsity and cold start problems. However, the existing methods cannot simultaneously acquire the feature information and structure information of the knowledge graph. In this paper, we propose RKG, a multi-task semantic feature and high-order structure learning approach for knowledge graph assisting recommendation. RKG, which consists of a Recommender module, a Knowledge graph feature learning module and a knowledge Graph structure learning module, is a deep end-to-end framework that utilizes knowledge graph’s latent information to enhance the recommender’s performance. Information is shared among the three modules through cross unit and exchange unit to interact automatically with multiple learning tasks. Extensive experiments on three public datasets show that RKG has achieved superior and stable performance in top-K recommendation, click-through rate prediction and sparse user-item interaction scenarios over several state-of-the-art baselines.

The performance of deep learning methods for person re-identification (Re-ID) is influenced by overfitting problem. To improve the generalization ability, most methods pay attention to generating and utilizing new samples. However, generated samples focus on object occlusion but neglect pedestrian occlusion, while triplet loss fails to preserve all identity information on samples with multiple pedestrians. In this paper, we propose the CutMix dual branch network (CDBN) to relieve these problems. CutMix is introduced to this model, and it is responsible for generating new samples with pedestrian occlusion. CutMix is verified complementary to image erasing strategy for Re-ID. Besides, a generalized triplet loss called CutMix triplet loss (CTP) is employed on samples augmented by CutMix with consideration of identity information from multiple pedestrians, making CDBN robust to two kinds of occlusions. Extensive experiments on two bench-marks demonstrate the strengths of CDBN, which is superior to state-of-the-art methods.

In natural language inference (NLI), attention mechanism has achieved great success, but it does not explicitly model order information between sequential elements. Besides, existing models for NLI tend to neglect positional information of inter-sentence words. In this paper, we propose a novel relative position representation over interaction space, which is called Distance and Direction based Relative Position Presentation (D2RPR). It can capture relative distance and relative direction information simultaneously, but also establish potential correlations between non-central words while representing the position of the central word. We incorporate D2RPR into self-attention over space to enhance intra-sentence contextual connections. Moreover, D2RPR is used to capture relative position relationships between inter-sentence words. In this way, our model can strengthen the semantic connection be-tween sentences. Experiment results show the effectiveness of D2RPR and our model achieves competitive performance on SNLI dataset and Quora dataset.

The increasing congestion of the RF spectrum is a key challenge for low-power wireless networks using concurrent transmissions. The presence of radio interference can indeed undermine their dependability, as they rely on a tight synchronization and incur a significant overhead to overcome packet loss. In this paper, we present Harmony, a new data collection protocol that exploits the benefits of concurrent transmissions and embeds techniques to ensure a reliable and timely packet delivery despite highly congested channels. Such techniques include, among others, a data freezing mechanism that allows to successfully deliver data in a partitioned network as well as the use of network coding to shorten the length of packets and increase the robustness to unreliable links. Harmony also introduces a distributed interference detection scheme that allows each node to activate various interference mitigation techniques only when strictly necessary, avoiding unnecessary energy expenditures while finding a good balance between reliability and timeliness. An experimental evaluation on real-world testbeds shows that Harmony outperforms state-of-the-art protocols in the presence of harsh Wi-Fi interference, with up to 50% higher delivery rates and significantly shorter end-to-end latencies, even when transmitting large packets.

This paper presents a real-time monocular SLAM algorithm which combines points and line segments. We extend traditional point-based SLAM system with line features which are usually abundant in man-made scenes. The system is more robust and accurate than traditional point-based and direct-based monocular SLAM algorithms. In order to improve the timeliness of multi-feature based SLAM, we propose a novel feature level parallel processing framework and a fast line matching algorithm. For improving the reconstruction accuracy of 3D line segments which is usually affected by unreliable line endpoints, a sample point-based 3D reconstruction algorithm for line segments is proposed. Our system is implemented based on a popular monocular SLAM known as ORB-SLAM and tested on the TUM RGB-D benchmark. The experiment results demonstrate that the proposed system performs better than current state-of-the-art visual SLAM with respect to accuracy.

This paper introduces our solution for iQIYI Celebrity Video Identification Challenge. After analyzing the iQIYI-VID-2019 dataset, we find the distribution of the dataset is very unbalanced and there are many unlabeled samples in the validation set and the test set. For these challenge, we propose a hierarchical system which combines different models and fuses base classifiers. For the false detections and low-quality features in the dataset, we use a simple and reasonable strategy to fuse features. In order to detect videos more accurately, we choose an improved loss function for the learning of base classifiers. Experiment results show that our framework performs well and evaluation conducted by the organizers shows that our final result gets the ninth place online and mAP 88.08%.

Based on pattern uniformity measure and the number of ones in the Local Binary Pattern (LBP) codes, this paper proposes an Improved Local Binary Pattern (ILBP) operator to describe local image texture more effectively. The ILBP operator discovers an important group of basic primitives such as lines, T-junctions, and cross-intersections, which are ignored by uniform LBP operator. Such local primitives are as crucial as those represented by uniform patterns for recognition tasks. The resulting ILBP feature is more discriminative than traditional LBP feature although they are both invariant in terms of monotonic gray-scale variation and rotation transformation.

We propose a heterogeneous, mid-level feature based method for recognizing natural scene categories. The proposed feature introduces spatial information among the latent topics by means of spatial pyramid, while the latent topics are obtained by using probabilistic latent semantic analysis (pLSA) based on the bag-of-words representation. The proposed feature always performs better than standard pLSA because the performance of pLSA is adversely affected in many cases due to the loss of spatial information. By combining various interest point detectors and local region descriptors used in the bag-of-words model, the proposed feature can make further improvement for diverse scene category recognition tasks. We also propose a two-stage framework for multi-class classification. In the first stage, for each of possible detector/descriptor pairs, adaptive boosting classifiers are employed to select the most discriminative topics and further compute posterior probabilities of an unknown image from those selected topics. The second stage uses the prod-max rule to combine information coming from multiple sources and assigns the unknown image to the scene category with the highest ‘final’ posterior probability. Experimental results on three benchmark scene datasets show that the proposed method exceeds most state-of-the-art methods.