Full-image dependencies provide useful contextual information to benefit visual understanding problems. In this work, we propose a Criss-Cross Network (CCNet) for obtaining such in more effective and efficient way. Concretely, each pixel, novel criss-cross attention module CCNet harvests the of all pixels on its path. By taking further recurrent operation, pixel can finally capture full-image from pixels. Overall, is with following merits: 1) GPU memory friendly. Compared non-local block,...

Pixel-level labelling tasks, such as semantic segmentation, play a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning techniques for recognition tackle pixel-level tasks. One issue this methodology is limited capacity delineate visual objects. To solve problem, we introduce new form convolutional neural network that combines strengths Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic...

While deep convolutional neural networks (CNNs) have shown a great success in single-label image classification, it is important to note that real world images generally contain multiple labels, which could correspond different objects, scenes, actions and attributes an image. Traditional approaches multi-label classification learn independent classifiers for each category employ ranking or thresholding on the results. These techniques, although working well, fail explicitly exploit label...

Letting a deep network be aware of the quality its own predictions is an interesting yet important problem. In task instance segmentation, confidence classification used as mask score in most segmentation frameworks. However, quality, quantified IoU between and ground truth, usually not well correlated with score. this paper, we study problem propose Mask Scoring R-CNN which contains block to learn predicted masks. The proposed takes feature corresponding together regress IoU. scoring...

We propose a learning-based hierarchical approach of multi-target tracking from single camera by progressively associating detection responses into longer and track fragments (tracklets) finally the desired target trajectories. To define tracklet affinity for association, most previous work relies on heuristically selected parametric models; while our is able to automatically select among various features corresponding non-parametric models, combine them maximize discriminative power...

While feedforward deep convolutional neural networks (CNNs) have been a great success in computer vision, it is important to note that the human visual cortex generally contains more feedback than connections. In this paper, we will briefly introduce background of feedbacks cortex, which motivates us develop computational mechanism networks. addition inference traditional networks, loop introduced infer activation status hidden layer neurons according "goal" network, e.g., high-level...

The recent development in learning deep representations has demonstrated its wide applications traditional vision tasks like classification and detection. However, there been little investigation on how we could build up a framework weakly supervised setting. In this paper, attempt to model (multiple instance learning) framework. our setting, each image follows dual multi-instance assumption, where object proposals possible text annotations can be regarded as two sets. We thus design...

Rotation invariant multiview face detection (MVFD) aims to detect faces with arbitrary rotation-in-plane (RIP) and rotation-off-plane (ROP) angles in still images or video sequences. MVFD is crucial as the first step automatic processing for general applications since are seldom upright frontal unless they taken cooperatively. In this paper, we propose a series of innovative methods construct high-performance rotation detector, including Width-First-Search (WFS) tree detector structure,...

The first global night-time light composite data from the Visible Infrared Imaging Radiometer Suite (VIIRS) day–night band carried by Suomi National Polar-orbiting Partnership (NPP) satellite were released recently. So far, few studies have been conducted to assess ability of NPP-VIIRS extract built-up urban areas. This letter aims evaluate potential this new-generation for extracting areas and compares results with Defense Meteorological Satellite Program–Operational Linescan System...

We present an approach for online learning of discriminative appearance models robust multi-target tracking in a crowded scene from single camera. Although much progress has been made developing methods optimal data association, there comparatively less work on the models, which are key elements good performance. Many previous either use simple features such as color histograms, or focus discriminability between target and background does not resolve ambiguities different targets. propose...

Poverty has appeared as one of the long-term predicaments facing development human society during 21st century. Estimation regional poverty level is a key issue for making strategies to eliminate poverty. This paper aims evaluate ability nighttime light composite data from Visible Infrared Imaging Radiometer Suite (VIIRS) Day-Night Band (DNB) carried by Suomi National Polar-orbiting Partnership (NPP) Satellite in estimating at county China. Two major experiments are involved this study,...

Face Recognition has been studied for many decades. As opposed to traditional hand-crafted features such as LBP and HOG, much more sophisticated can be learned automatically by deep learning methods in a data-driven way. In this paper, we propose two-stage approach that combines multi-patch CNN metric learning, which extracts low dimensional but very discriminative face verification recognition. Experiments show method outperforms other state-of-the-art on LFW dataset, achieving 99.77%...

In this paper we examine the effect of receptive field designs on classification accuracy in commonly adopted pipeline image classification. While existing algorithms usually use manually defined spatial regions for pooling, show that learning more adaptive fields increases performance even with a significantly smaller codebook size at coding layer. To learn optimal pooling parameters, adopt idea over-completeness by starting large number candidates, and train classifier structured sparsity...

The prevalent scene text detection approach follows four sequential steps comprising character candidate detection, false removal, line extraction, and verification. However, errors occur accumulate throughout each of these which often lead to low performance. To address issues, we propose a unified system, namely Text Flow, by utilizing the minimum cost (min-cost) flow network model. With candidates detected cascade boosting, min-cost model integrates last three into single process solves...

This paper presents a data-driven matching cost for stereo matching. A novel deep visual correspondence embedding model is trained via Convolutional Neural Network on large set of images with ground truth disparities. leverages appearance data to learn similarity relationships between corresponding image patches, and explicitly maps intensity values into an feature space measure pixel dissimilarities. Experimental results KITTI Middlebury sets demonstrate the effectiveness our model. First,...

In this paper, we propose a conceptually novel, efficient, and fully convolutional framework for real-time instance segmentation. Previously, most segmentation methods heavily rely on object detection perform mask prediction based bounding boxes or dense centers. contrast, sparse set of activation maps, as new representation, to high-light informative regions each foreground object. Then instance-level features are obtained by aggregating according the highlighted recognition Moreover,...

Autonomous driving requires a comprehensive understanding of the surrounding environment for reliable trajectory planning. Previous works rely on dense rasterized scene representation (e.g., agent occupancy and semantic map) to perform planning, which is computationally intensive misses instance-level structure information. In this paper, we propose VAD, an end-to-end vectorized paradigm autonomous driving, models as fully representation. The proposed has two significant advantages. On one...

Recent advances in hydrological modling have led to the generation of numerous global or regional runoff datasets, which been widely used analysis. However, it is not yet clear how their accuracy and reliabilities are. In this study, using observed gauge streamflow data at four stations (Hequ, Fugu, Wubu, Longmen) middle reaches Yellow River as reference, we compare evaluate three gridded dataset products (GloFAS, GRFR v1.0, WGHM) temporal scales: daily, monthly, annual, wet/dry seasons. The...

In this paper, we propose a novel tree-structured multiview face detector (MVFD), which adopts the coarse-to-fine strategy to divide entire space into smaller and subspaces. For purpose, newly extended boosting algorithm named vector is developed train predictors for branching nodes of tree that have multicomponents outputs as vectors. Our MVFD covers large range space, say, +/-45/spl deg/ rotation in plane (RIP) +/-90/spl off (ROP), achieves high accuracy amazing speed (about 40 ms per...