Conventional methods for object detection typically require a substantial amount of training data and preparing such high-quality is very labor-intensive. In this paper, we propose novel few-shot network that aims at detecting objects unseen categories with only few annotated examples. Central to our method are Attention-RPN, Multi-Relation Detector Contrastive Training strategy, which exploit the similarity between shot support set query detect while suppressing false in background. To...

Self-training via pseudo labeling is a conventional, simple, and popular pipeline to leverage unlabeled data. In this work, we first construct strong baseline of self-training (namely ST) for semi-supervised semantic segmentation injecting data augmentations (SDA) on images alleviate overfitting noisy labels as well decouple similar predictions between the teacher student. With simple mechanism, our ST outperforms all existing methods without any bells whistles, e.g., iterative retraining....

We tackle the problem of single image depth estimation, which, without additional knowledge, suffers from many ambiguities. Unlike previous approaches that only reason locally, we propose to exploit global structure scene estimate its depth. To this end, introduce a hierarchical representation scene, which models local jointly with mid-level and structures. formulate estimation as inference in graphical model whose edges let us encode interactions within across different layers our...

Over the years, indoor scene parsing has attracted a growing interest in computer vision community. Existing methods have typically focused on diverse subtasks of this challenging problem. In particular, while some them aim at segmenting image into regions, such as object or surface instances, others inferring semantic labels given their support relationships. These different tasks are treated separate ones. However, they bear strong connections: good regions should respect labels, can only...

Compared to fully supervised object detection, training with sparse annotations typically leads a decline in performance due insufficient feature diversity. Existing sparsely annotated detection (SAOD) methods often rely on pseudo-labeling strategies, but these pseudo-labels tend introduce noise under extreme sparsity. To simultaneously avoid the impact of pseudo-label and enhance diversity, we propose novel Adaptive Feature Generation (AdaptFG) model that generates features based class...

Benefiting from the massive labeled samples, deep learning-based segmentation methods have achieved great success for two dimensional natural images. However, it is still a challenging task to segment high medical volumes and sequences, due considerable efforts clinical expertise make large scale annotations. Self/semi-supervised learning been shown improve performance by exploiting unlabeled data. they are lack of mining local semantic discrimination exploitation volume/sequence structures....

This paper presents MalwareVis, a utility that provides security researchers method to browse, filter, view and compare malware network traces as entities.

In this work, we present a fully self-supervised framework for semantic segmentation(FS^4). A bootstrapped strategy segmentation, which saves efforts the huge amount of annotation, is crucial building customized models from end-to-end open-world domains. This application eagerly needed in realistic scenarios. Even though recent segmentation methods have gained great progress, these works however heavily depend on fully-supervised pretrained model and make it impossible pipeline. To solve...

Class Activation Map (CAM) has emerged as a popular tool for weakly supervised semantic segmentation (WSSS), allowing the localization of object regions in an image using only image-level labels. However, existing CAM methods suffer from under-activation target and false-activation background due to fact that lack detailed supervision can hinder model's ability understand whole. In this paper, we propose novel Question-Answer Cross-Language-Image Matching framework WSSS (QA-CLIMS),...

Class Activation Map (CAM) has emerged as a popular tool for weakly supervised semantic segmentation (WSSS), allowing the localization of object regions in an image using only image-level labels. However, existing CAM methods suffer from under-activation target and false-activation background due to fact that lack detailed supervision can hinder model's ability understand whole. In this paper, we propose novel Question-Answer Cross-Language-Image Matching framework WSSS (QA-CLIMS),...

Despite the great progress of semantic segmentation with supervised learning, annotating large amounts pixel-wise labels is, however, very expensive and time-consuming. To this end, Unsupervised Semantic Segmentation(USS) has been proposed to learn segmentation, without any form annotations. This approach involves dense prediction semantics which is however challenging due unreliable nature local representations. solve problem, we propose a newly context-aware unsupervised framework, aims...

Conventional methods for object detection typically require a substantial amount of training data and preparing such high-quality is very labor-intensive. In this paper, we propose novel few-shot network that aims at detecting objects unseen categories with only few annotated examples. Central to our method are Attention-RPN, Multi-Relation Detector Contrastive Training strategy, which exploit the similarity between shot support set query detect while suppressing false in background. To...

Modern object detection methods typically rely on bounding box proposals as input. While initially popularized in the 2D case, this idea has received increasing attention for 3D boxes. Nevertheless, existing proposal techniques all assume having access to depth input, which is unfortunately not always available practice. In paper, we therefore introduce an approach generating from a single monocular RGB image. To end, develop integrated, fully differentiable framework that inherently...

In publishing and printing of network version field, there are enormous number TIFF format (CMYK) images which requires too huge space for storing enough bandwidth transmitting. Therefore, common need to manipulate amount data brought about the issue fast lossless compression. 2D integer wavelet transform can be used compression static image, such as, 5/3 lifting is JEPG2000. Today, Multi-core (dual, four or eight cores) CPU technology help accelerate speed. However, current multi-more limit...

Watching micro-videos is becoming a part of public daily life. Usually, user watching behaviors are thought to be rooted in their multiple different interests. In the paper, we propose model named OPAL for micro-video matching, which elicits user's heterogeneous interests by disentangling soft and hard interest embeddings from interactions. Moreover, employs two-stage training strategy, pre-train generate historical interactions under guidance orthogonal hyper-categories fine-tune reinforce...

Self-training via pseudo labeling is a conventional, simple, and popular pipeline to leverage unlabeled data. In this work, we first construct strong baseline of self-training (namely ST) for semi-supervised semantic segmentation injecting data augmentations (SDA) on images alleviate overfitting noisy labels as well decouple similar predictions between the teacher student. With simple mechanism, our ST outperforms all existing methods without any bells whistles, e.g., iterative re-training....

Network is widely used in today. However, limit bandwidth and no security are the current network situation. Transmission of image data Internet may be affected because above shortcomings. On one hand, transmission important must efficient secure, such as, medical images, remote sensing image, so on. other large images (TIFF, BMP, DCMOM), applied reality, which cause difficult, or unavailable Internet. Considering scheme, using FTP, QICQ, email general tools, however, they can't satisfy both...

Two non-destructive detection methods for potato blight based on hyperspectral imaging were used: convolutional neural network (CNN) and support vector machine (SVM) to classify leaves. By comparing the classification results, advantages disadvantages of different are analyzed. In experiment, normal leaves early selected as research objects. Hyperspectral images samples obtained by system, then principal component extracted analysis method. It was found that significantly different, finally...