Mainstream object detectors based on the fully convolutional network has achieved impressive performance. While most of them still need a hand-designed non-maximum suppression (NMS) post-processing, which impedes end-to-end training. In this paper, we give analysis discarding NMS, where results reveal that proper label assignment plays crucial role. To end, for detectors, introduce Prediction-aware One-To-One (POTO) classification to enable detection, obtains comparable performance with NMS....

Recently, numerous handcrafted and searched networks have been applied for semantic segmentation. However, previous works intend to handle inputs with various scales in pre-defined static architectures, such as FCN, U-Net, DeepLab series. This paper studies a conceptually new method alleviate the scale variance representation, named dynamic routing. The proposed framework generates data-dependent routes, adapting distribution of each image. To this end, differentiable gating function, called...

Current state-of-the-art approaches for spatio-temporal action detection have achieved impressive results but remain unsatisfactory temporal extent detection. The main reason comes from that, there are some ambiguous states similar to the real actions which may be treated as target even by a well trained network. In this paper, we define these samples "transitional states", and propose Transition-Aware Context Network (TACNet) distinguish transitional states. proposed TACNet includes two...

In this paper, we address the challenging problem of weakly supervised spatio-temporal action localization for which only video-level labels are available during training. To solve problem, propose an end-to-end Global Local Network (GLNet) to predict probability distribution simultaneously in both spatial and temporal space. The proposed GLNet model includes two key components: a local module global module. aims frame-level by encoding short-term information. particular, Region Actionness...

The Feature Pyramid Network (FPN) presents a remarkable approach to alleviate the scale variance in object representation by performing instance-level assignments. Nevertheless, this strategy ignores distinct characteristics of different sub-regions an instance. To end, we propose fine-grained dynamic head conditionally select pixel-level combination FPN features from scales for each instance, which further releases ability multi-scale feature representation. Moreover, design spatial gate...

The timeliness of annual reports has received serious attention by the Chinese regulatory and professional bodies in recent years. Using 8294 company from listed companies during 1993–2003, this paper analyzes first examining trend reporting lags, separating effects 'good' 'bad' news on timeliness, then using a multivariate regression to identify determinants lags. It is found that there significant improvement although shortening lags shows U-shape over data period. There strong evidence...

Learning discriminative global features plays a vital role in semantic segmentation. And most of the existing methods adopt stacks local convolutions or non-local blocks to capture long-range context. However, due absence spatial structure preservation, these operators ignore object details when enlarging receptive fields. In this paper, we propose learnable tree filter form generic filtering module that leverages structural property minimal spanning model dependencies while preserving...

China has been making efforts in nature conservation by developing a new national park system. Setting fee-based entrance policy for the newly established parks can be challenging without information on visitors' willingness to pay (WTP). Thus, this study aims evaluate WTP fees with dataset of 1215 visitors collected China's planned Qinling National Park (QNP). Using double bounded dichotomous choice format contingent valuation method, we obtained mean fee QNP 200 yuan. Visitors' demand...

Large disparity stereo matching is critical to the application of a vision system especially for outdoor scenes. Nevertheless, how efficiently design high accuracy large-disparity on field-programmable gate array (FPGA) still grand challenge. The computational complexity previously proposed inevitably proportional range; hence their hardware designs become very inefficient when range large. Motivated by original PatchMatch and weighted median filtering (WMF) algorithms, this paper proposes...

Excessive support from banks and governments to poorly‐performing firms, making them zombie distorts financial resources—a common factor in firm innovation. Using a large sample of Chinese industrial this paper investigates the impact firms on innovation healthy firms. The results indicate that prevalence reduces patent output seriously jeopardises Endogeneity analyses robustness checks validate results. Further analysis reveals distort credit resources especially those are state‐owned or...

Abstract Background Transversus abdominis plane (TAP) block can provide effective analgesia for abdominal surgery. However, it was questionable whether TAP had additional effect in the context of multimodal (MMA). Therefore, this study aimed to assess analgesic preoperative when added MMA protocol open gynecological Methods In prospective, randomized-controlled trial, 64 patients scheduled surgery were randomized receive (Study group, n = 32) or placebo (Control addition comprising...

The Learnable Tree Filter presents a remarkable approach to model structure-preserving relations for semantic segmentation. Nevertheless, the intrinsic geometric constraint forces it focus on regions with close spatial distance, hindering effective long-range interactions. To relax constraint, we give analysis by reformulating as Markov Random Field and introduce learnable unary term. Besides, propose spanning tree algorithm replace original non-differentiable one, which further improves...

Face detection has achieved tremendous strides thanks to convolutional neural networks. However, dense face remains an open challenge due large scale variation, tiny faces, and serious occlusion. This paper presents a robust, detector using global context visual attention mechanisms which can significantly improve accuracy. Specifically, fusion module with top-down feedback is proposed the ability identify faces. Moreover, mechanism employed solve problem of Experimental results on public...

Objective This study aimed to develop a deep learning system identify and differentiate the metastatic cervical lymph nodes (CLNs) of thyroid cancer. Methods From January 2014 December 2020, 3059 consecutive patients with suspected CLNs cancer were retrospectively enrolled in this study. All confirmed by fine needle aspiration. The randomly divided into training (1228 benign 1284 CLNs) test (307 240 groups. Grayscale ultrasonic images used performance Y-Net model. We network model segment...

Microplastics (MPs) have been found in both surface water and groundwater, which are sources of drinking water.Since one the most important routes for MPs to enter human body is through water, enormous buildup waterbodies resulting effects on health caused social concern.However, our knowledge treatment plant (DWTP) techniques affects removal MPs, there aren't any standardized or efficient quality assurance control (QA/QC) measures sampling analysis.The current state described this review,...

Recently, numerous handcrafted and searched networks have been applied for semantic segmentation. However, previous works intend to handle inputs with various scales in pre-defined static architectures, such as FCN, U-Net, DeepLab series. This paper studies a conceptually new method alleviate the scale variance representation, named dynamic routing. The proposed framework generates data-dependent routes, adapting distribution of each image. To this end, differentiable gating function, called...

Sample classification, especially disease status prediction, is an important area of investigation for gene expression studies. Many machine learning methods have been developed to tackle this problem. To evaluate different prediction methods, the IMPROVER Challenge made several data sets available. Here we focus on one sub-challenge: chronic obstructive pulmonary (COPD). We outlined critical preprocessing steps make training and test comparable. compared our recently introduced random...

Mainstream object detectors based on the fully convolutional network has achieved impressive performance. While most of them still need a hand-designed non-maximum suppression (NMS) post-processing, which impedes end-to-end training. In this paper, we give analysis discarding NMS, where results reveal that proper label assignment plays crucial role. To end, for detectors, introduce Prediction-aware One-To-One (POTO) classification to enable detection, obtains comparable performance with NMS....

In this report, we introduce our real-time 2D object detection system for the realistic autonomous driving scenario. Our detector is built on a newly designed YOLO model, called YOLOX. On Argoverse-HD dataset, achieves 41.0 streaming AP, which surpassed second place by 7.8/6.1 detection-only track/fully track, respectively. Moreover, equipped with TensorRT, model 30FPS inference speed high-resolution input size (e.g., 1440-2304). Code and models will be available at...

Feature selection is often required to select a feature subset from the original set of objects very high resolution (VHR) remote sensing images. However, majority methods supervised, and could fail identify relevant features when labeled are scarce. To address problem, this paper proposes method, efficient semi-supervised (ESFS), by effectively exploiting underlying information huge amount unlabeled objects. Firstly, probability matrix utilized in loss function measure relevance on classes,...

Current state-of-the-art approaches for spatio-temporal action detection have achieved impressive results but remain unsatisfactory temporal extent detection. The main reason comes from that, there are some ambiguous states similar to the real actions which may be treated as target even by a well-trained network. In this paper, we define these samples "transitional states", and propose Transition-Aware Context Network (TACNet) distinguish transitional states. proposed TACNet includes two...