Chest X-ray film is the most widely used and common method of clinical examination for pulmonary nodules. However, number radiologists obviously cannot keep up with this outburst due to sharp increase in diseases, which increases rate missed diagnosis misdiagnosis. The based on deep learning appropriate way deal such problems so far. main research paper was using inception-v3 transfer model classify images, finally get a practical feasible computer-aided diagnostic model. could improve...

The high cost of pixel-level annotations makes it appealing to train saliency detection models with weak supervision. However, a single supervision source usually does not contain enough information well-performing model. To this end, we propose unified framework diverse sources. In paper, use category labels, captions, and unlabelled data for training, yet other sources can also be plugged into flexible framework. We design classification network (CNet) caption generation (PNet), which...

Existing weakly supervised semantic segmentation (WSSS) methods usually utilize the results of pre-trained saliency detection (SD) models without explicitly modelling connections between two tasks, which is not most efficient configuration. Here we propose a unified multi-task learning framework to jointly solve WSSS and SD using single network, i.e. network (SSNet). SSNet consists (SN) aggregation module (SAM). For an input image, SN generates result and, SAM predicts each category...

Recent deep generative inpainting methods use attention layers to allow the generator explicitly borrow feature patches from known region complete a missing region. Due lack of supervision signals for correspondence between regions and regions, it may fail find proper reference features, which often leads artifacts in results. Also, computes pair-wise similarity across entire map during inference bringing significant computational overhead. To address this issue, we propose teach such...

The categories and appearance of salient objects vary from image to image, therefore, saliency detection is an image-specific task. Due lack large-scale training data, using deep neural networks (DNNs) with pretraining difficult precisely capture the cues. To solve this issue, we formulate a zero-shot learning problem promote existing detectors. Concretely, DNN trained as embedding function map pixels attributes salient/background regions into same metric space, in which classifier learned...

Sketch-based image manipulation is an interactive editing task to modify based on input sketches from users. Existing methods typically formulate this as a conditional inpainting problem, which requires users draw extra mask indicating the region in addition sketches. The masked regions are regarded holes and filled by model conditioned sketch. With formulation, paired training data can be easily obtained randomly creating masks extracting edges or contours. Although setup simplifies...

Accurate segmentation of lesions is crucial for diagnosis and treatment early esophageal cancer (EEC). However, neither traditional nor deep learning-based methods up to today can meet the clinical requirements, with mean Dice score - most important metric in medical image analysis hardly exceeding 0.75. In this paper, we present a novel learning approach segmenting EEC lesions. Our method stands out its uniqueness, as it relies solely on single input from patient, forming so-called...

We propose a novel unsupervised game-theoretic salient object detection algorithm that does not require labeled training data. First, saliency problem is formulated as non-cooperative game, hereinafter referred to Saliency Game, in which image regions are players who choose be "background" or "foreground" their pure strategies. A payoff function constructed by exploiting multiple cues and combining complementary features. maps generated according each region's strategy the Nash equilibrium...

Existing weakly supervised semantic segmentation (WSSS) methods usually utilize the results of pre-trained saliency detection (SD) models without explicitly modeling connections between two tasks, which is not most efficient configuration. Here we propose a unified multi-task learning framework to jointly solve WSSS and SD using single network, \ie saliency, network (SSNet). SSNet consists (SN) aggregation module (SAM). For an input image, SN generates result and, SAM predicts each category...

Benefiting from the rapid development of Convolutional Neural Networks (CNNs), some salient object detection methods have achieved remarkable results by utilizing multi-level convolutional features. However, saliency training datasets is limited scale due to high cost pixel-level labeling, which leads a generalization trained model on new scenarios during testing. Besides, FCN-based directly integrate features, ignoring fact that noise in features are harmful detection. In this paper, we...

High-cost pixel-level annotations makes it appealing to train saliency detection models with weak supervision. However, a single supervision source hardly contain enough information well-performing model. To this end, we introduce unified two-stage framework learn from category labels, captions, web images and unlabeled images. In the first stage, design classification network (CNet) caption generation (PNet), which predict object categories generate respectively, meanwhile highlights...

By considering different weights of items, weighted frequent pattern (WFP) mining can find more important patterns. However previous WFP algorithms are not suitable for continuous, unbounded and high-speed data streams they need multiple database scans. In this paper, we present an efficient algorithm DSWFP, which is based on sliding window discover from the recent data. DSWFP has three new characters, including a refined weight definition, proposed structure two pruning strategies....

Here, we explore the low-level statistics of images generated by state-of-the-art deep generative models. First, Variational auto-encoder (VAE~\cite{kingma2013auto}), Wasserstein adversarial network (WGAN~\cite{arjovsky2017wasserstein}) and convolutional (DCGAN~\cite{radford2015unsupervised}) are trained on ImageNet dataset a large set cartoon frames from animations. Then, for these models as well natural scenes cartoons, including mean power spectrum, number connected components in given...

Carrier sense multiple access with collision notification (CSMA/CN) is a typical representation of physical-layer (PHY)/medium control (MAC) co-designs, where the MAC frames are implemented or detected using PHY techniques. With CSMA/CN, sender detects an unsuccessful transmission, aid (CN) from receiver. In this paper, we first theoretically study crucial impact CN attributes (namely, detection threshold and length) on system throughput in wireless local area network (WLAN). We identify...

In this paper, an obstacle detection and avoidance system based on laser radar for mobile robots is established, a new integrated improved algorithm proposed. The composed of industrial computer, robot. mounted the robot, environmental information in front robot obtained through sent to computer. Set scrolling window layer adopt different control strategies levels obstacles. experimental results show that can effectively detect avoid

Face reenactment and reconstruction benefit various applications in self-media, VR, etc. Recent face methods use 2D facial landmarks to implicitly retarget expressions poses from driving videos source images, while they suffer pose expression preservation issues for cross-identity scenarios, i.e., when the subjects are different. Current self-supervised also demonstrate impressive results. However, these do not handle large well, since their training data lacks samples of expressions,...

At the core of portrait photography is search for ideal lighting and viewpoint. The process often requires advanced knowledge in an elaborate studio setup. In this work, we propose Holo-Relighting, a volumetric relighting method that capable synthesizing novel viewpoints, from single image. Holo-Relighting leverages pretrained 3D GAN (EG3D) to reconstruct geometry appearance input as set 3D-aware features. We design module conditioned on given these features, predict relit representation...

Visible-infrared person re-identification (VI-reID) aims at matching cross-modality pedestrian images captured by disjoint visible or infrared cameras. Existing methods alleviate the discrepancies via designing different kinds of network architectures. Different from available methods, in this paper, we propose a novel parameter optimizing paradigm, hierarchical optimization (PHO) method, for task VI-ReID. It allows part parameters to be directly optimized without any training, which narrows...

Personalized text-to-image generation models enable users to create images that depict their individual possessions in diverse scenes, finding applications various domains. To achieve the personalization capability, existing methods rely on finetuning a foundation model user's custom dataset, which can be non-trivial for general users, resource-intensive, and time-consuming. Despite attempts develop finetuning-free methods, quality is much lower compared counterparts. In this paper, we...

Motivation: Arterial spin labeling (ASL) is vulnerable to motion and off-resonance, which may result in unstable image quality, particularly children. Goal(s): To propose an automatic quality assessment model for 3D ASL Approach: The proposed was trained validated on 51 scans from children, compared a previously developed reference method. performance evaluated using AUC 5-fold cross-validation tests. Results: yielded 8%-11% higher AUC, accuracy, F1 score the Impact: assessing children's...