Person re-identification (re-ID) models trained on one domain often fail to generalize well another. In our attempt, we present a "learning via translation" framework. the baseline, translate labeled images from source target in an unsupervised manner. We then train re-ID with translated by supervised methods. Yet, being essential part of this framework, image-image translation suffers information loss source-domain labels during translation. Our motivation is two-fold. First, for each...

Within Convolutional Neural Network (CNN), the convolution operations are good at extracting local features but experience difficulty to capture global representations. visual transformer, cascaded self-attention modules can long-distance feature dependencies unfortunately deteriorate details. In this paper, we propose a hybrid network structure, termed Conformer, take advantage of convolutional and mechanisms for enhanced representation learning. Conformer roots in Feature Coupling Unit...

Detecting objects in aerial images is challenged by variance of object colors, aspect ratios, cluttered backgrounds, and particular, undetermined orientations. In this paper, we propose to use Deep Convolutional Neural Network (DCNN) features from combined layers perform orientation robust detection. We explore the inherent characteristics DC-NN as well relate extracted principle disentangling feature learning. An image segmentation based approach used localize ROIs various are further...

Weakly supervised instance segmentation with image-level labels, instead of expensive pixel-level masks, remains unexplored. In this paper, we tackle challenging problem by exploiting class peak responses to enable a classification network for mask extraction. With image labels supervision only, CNN classifiers in fully convolutional manner can produce response maps, which specify confidence at each location. We observed that local maximums, i.e., peaks, map typically correspond strong...

In this paper, we present a large-scale dataset and establish baseline for prohibited item discovery in Security Inspection X-ray images. Our dataset, named SIXray, consists of 1,059,231 images, which 6 classes 8,929 items are manually annotated. It raises brand new challenge overlapping image data, meanwhile shares the same properties with existing datasets, including complex yet meaningless contexts class imbalance. We propose an approach class-balanced hierarchical refinement (CHR) to...

Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling significant local and global rotations remains limited. In this paper, we propose Active Rotating Filters (ARFs) that actively rotate during convolution produce feature maps with location orientation explicitly encoded. An ARF acts as a virtual filter bank containing the itself its multiple unmaterialised rotated versions. During...

Weakly supervised object detection (WSOD) is a challenging task when provided with image category supervision but required to simultaneously learn locations and detectors. Many WSOD approaches adopt multiple instance learning (MIL) have non-convex loss functions which are prone get stuck into local minima (falsely localize parts) while missing full extent during training. In this paper, we introduce continuation optimization method MIL thereby creating (C-MIL), the intention of alleviating...

Fine-grained recognition poses the unique challenge of capturing subtle inter-class differences under considerable intra-class variances (e.g., beaks for bird species). Conventional approaches crop local regions and learn detailed representation from those regions, but suffer fixed number parts missing surrounding context. In this paper, we propose a simple yet effective framework, called Selective Sparse Sampling, to capture diverse fine-grained details. The framework is implemented using...

Weakly supervised object detection is a challenging task when provided with image category supervision but required to learn, at the same time, locations and detectors. The inconsistency between weak learning objectives introduces randomness ambiguity In this paper, min-entropy latent model (MELM) proposed for weakly detection. Min-entropy used as metric measure of localization during learning, well serving learn locations. It aims principally reduce variance positive instances alleviate...

Weakly supervised object localization remains a challenge when learning models from image category labels. Optimizing classification tends to activate parts and ignore the full extent, while expanding into extent could deteriorate performance of classification. In this paper, we propose divergent activation (DA) approach, target at complementary discriminative visual patterns for weakly perspective discrepancy. To end, design hierarchical (HDA), which leverages semantic discrepancy spread...

Detecting oriented and densely packed objects remains challenging for spatial feature aliasing caused by the intersection of reception fields between objects. In this paper, we propose a convex-hull adaptation (CFA) approach configuring convolutional features in accordance with object layouts. CFA is rooted representation, which defines set dynamically predicted points guided convex over union (CIoU) to bound extent pursues optimal assignment constructing sets splitting positive or negative...

With convolution operations, Convolutional Neural Networks (CNNs) are good at extracting local features but experience difficulty to capture global representations. cascaded self-attention modules, vision transformers can long-distance feature dependencies unfortunately deteriorate details. In this paper, we propose a hybrid network structure, termed Conformer, take both advantages of operations and mechanisms for enhanced representation learning. Conformer roots in coupling CNN transformer...

Flexible and crack-free bacterial cellulose (BC)–silica composite aerogels (CAs) are prepared through a sol–gel process followed by freeze drying, in which the BC matrix silica gel skeleton form an interpenetrating network microstructure. The BC–silica CAs exhibit low density (0.02 g cm−3), high specific surface area (734.1 m2 g−1) thermal conductivity (0.031 W m−1 K−1), almost same as pure aerogels. Due to synergic effects of skeleton, obtained show excellent robustness flexibility overcome...

Weakly supervised object localization remains challenging, where only image labels instead of bounding boxes are available during training. Object proposal is an effective component in localization, but often computationally expensive and incapable joint optimization with some the remaining modules. In this paper, to best our knowledge, we for first time integrate weakly into convolutional neural networks (CNNs) end-to-end learning manner. We design a network component, Soft Proposal (SP),...

In this paper, we establish a baseline for object symmetry detection in complex backgrounds by presenting new benchmark and an end-to-end deep learning approach, opening up promising direction the wild. The benchmark, named Sym-PASCAL, spans challenges including diversity, multi-objects, part-invisibility, various that are far beyond those existing datasets. proposed Side-output Residual Network (SRN), leverages output Units (RUs) to fit errors between ground-truth outputs of RUs. By...

Weakly supervised object detection is a challenging task when provided with image category supervision but required to learn, at the same time, locations and detectors. The inconsistency between weak learning objectives introduces significant randomness ambiguity In this paper, min-entropy latent model (MELM) proposed for weakly detection. Min-entropy serves as learn metric measure of localization during learning. It aims principally reduce variance learned instances alleviate MELM...

Discriminative region responses residing inside an object instance can be extracted from networks trained with image-level label supervision. However, learning the full extent of pixel-level response in a weakly supervised manner remains unexplored. In this work, we tackle challenging problem by using novel filling approach. We first design process to selectively collect pseudo supervision noisy segment proposals obtained previously published techniques. The is used learn differentiable...

The gap in data distribution motivates domain adaptation research. In this area, image classification intrinsically requires the source and target features to be co-located if they are of same class. However, many works only take a global view gap. That is, make distributions globally overlap; does not necessarily lead feature co-location at class level. To resolve problem, we study metric learning context adaptation. Specifically, introduce similarity guided constraint (SGC)....

Unmanned Aerial Vehicles (UAV) have many applications in both commerce and recreation. However, irresponsibly operated UAVs will pose a threat to public safety. Therefore, developing our understanding of their uses is particular interest. This paper considers tracking UAVs, which provide multifaceted information around location, paths trajectories. To facilitate research on this topic, we introduce new benchmark, herein referred as Anti-UAV, provides novel direction for UAV with more than...

Bacterial cellulose (BC)–silica composite aerogels (CAs) with interpenetrating network (IPN) microstructure are prepared through a permeation sol–gel process followed by freeze drying. The IPN structure is constructed diffusing the precursor into three-dimensional (3D) BC matrix permeating catalyst gradually to promote in situ condensation of form SiO2 gel skeleton from outside inside. used here Na2SiO3 instead traditional tetraethoxysilane. This could offer excellent mechanical properties...

Unlike weld seam detection in a welding process, line localization for inspection is usually performed outdoors and challenged by noise variation of illumination intensity. In this paper, we propose approach mobile platform via cross structured light (CSL) device spatial-temporal cascaded hidden Markov models (HMMs). A CSL designed to project red laser stripes on weldment surfaces capture the convexity video sequences. Stripe edge images are extracted then spatial HMM detect regions interest...