The unpaired image-to-image translation aims to translate input images from one source domain some desired outputs in a target by learning training data. Cycle-consistency constraint provides general principle estimate and measure forward backward mapping functions between two domains. In many cases, the information entropy of domains is not equal, resulting an information-rich information-poor domain. However, existing solutions based on cycle-consistency either completely discard asymmetry...

Visual images corrupted by various types and levels of degradations are commonly encountered in practical image compression. However, most existing compression methods tailored for clean images, therefore struggling to achieve satisfying results on these images. Joint restoration typically focus a single type degradation fail address variety practice. To this end, we propose unified framework all-in-one restoration, which incorporates the capability against into process The key challenges...

The huge domain gap between sketches and photos the highly abstract sketch representations pose challenges for sketch-based image retrieval (\underline{SBIR}). zero-shot (\underline{ZS-SBIR}) is more generic practical but poses an even greater challenge because of additional knowledge seen unseen categories. To simultaneously mitigate both gaps, we propose \textbf{A}pproaching-and-\textbf{C}entralizing \textbf{Net}work (termed "\textbf{ACNet}") to jointly optimize sketch-to-photo synthesis...

Chinese word embeddings have recently attracted much attention in natural language processing (NLP). Existing researches learn based on characters, radicals, components and stroke n-gram. Besides abovementioned features, characters also own structure pinyin features. In this paper, we design feature substring, a super set of n-gram with information, to integrate stroke, features capture the semantics words. Based propose novel method ssp2vec predict contextual words substrings target for...

As an effective and novel knowledge management technology, graph can provide a new way for the inheritance development of traditional Chinese medicine (TCM). However, construction TCM is still mainly based on structured data at present. With accumulation literatures electronic medical records, large amount stored in unstructured texts which urgently needs to be extracted learning. In this study, we extract core concepts build ontology layer by analyzing process diagnosis treatment. Then use...

Low illumination, light reflections, scattering, absorption, and suspended particles inevitably lead to critically degraded underwater image quality, which poses great challenges for recognizing objects from images. The existing enhancement methods that aim promote visibility heavily suffer poor restoration performance generalization ability. To reduce the difficulty of enhancement, we introduce media transmission map as guidance enhancement. Different frameworks, also medium better...

The existing face forgery algorithms have achieved remarkable progress in how to generate reasonable facial images and can even successfully deceive human beings. Considering public security, detection is of vital importance, making it essential design detect over the Internet. Despite great success by Deepfake algorithms, they usually failed achieve satisfactory performance when deployed handle videos practice. One significant reason compression. Internet are inevitably compressed...

Recent research in cross-domain image retrieval has focused on addressing two challenging issues: handling domain variations the data and dealing with lack of sufficient training labels. However, these problems have often been studied separately, limiting practicality significance outcomes. The existing setting is also restricted to cases where labels are known during training, all samples semantic category information or instance correspondences. In this paper, we propose a novel approach...

Steganography is an important and prevailing information hiding tool to perform secret message transmission in open environment. Existing steganography methods can mainly fall into two categories: predefined rule-based data-driven methods. The former susceptible the statistical attack, while latter adopts deep convolution neural networks promote security. However, learning-based suffer from perceptible artificial artifacts or steganalysis. In this article, we introduce a novel...

Marine object detection has gained prominence in marine research, driven by the pressing need to unravel oceanic mysteries and enhance our understanding of invaluable ecosystems. There is a profound requirement efficiently accurately identify localize diverse unseen entities within underwater imagery. The open-marine (OMOD for short) required detect objects, performing categorization localization simultaneously. To achieve OMOD, we present \textbf{MarineDet}. We formulate joint visual-text...

This paper addresses an important and valuable open-world object detection (OWOD) in autonomous driving scenarios, which aims to detect objects under both <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">domain-agnostic</i> xmlns:xlink="http://www.w3.org/1999/xlink">category-agnostic</i> settings simultaneously. Existing OWOD algorithms mainly focus on the of pre-defined categories various conditions (domain-agnostic) or instead perform...

Coral reefs formulate the most valuable and productive marine ecosystems, providing habitat for many species. reef surveying analysis are currently confined to coral experts who invest substantial effort in generating comprehensive dependable reports (\emph{e.g.}, coverage, population, spatial distribution, \textit{etc}), from collected survey data. However, performing dense based on manual efforts is significantly time-consuming, existing algorithms compromise opt down-sampling only...

Existing domain adaptive object detection algorithms (DAOD) have demonstrated their effectiveness in discriminating and localizing objects across scenarios. However, these typically assume a single source target for adaptation, which is not representative of the more complex data distributions practice. To address this issue, we propose novel Open-Scenario Domain Adaptive Object Detection (OSDA), leverages multiple domains practical effective adaptation. We are first to increase granularity...

With the rapid development of Internet technology, scale knowledge and data production continues to expand. Such huge has brought convenience our lives challenges research. In order obtain quickly accurately, this paper uses machine learning algorithms build a multidimensional search model. The model adds real-time mechanism acquire various knowledge, so that system intelligent features such as self-learning selfadaptation. Through comparative analysis traditional algorithms, results show...

Addressing domain shifts for complex perception tasks in autonomous driving has long been a challenging problem. In this paper, we show that existing adaptation methods pay little attention to the content mismatch issue between source and target domains, thus weakening per-formance decoupling of domain-invariant domain-specific representations. To solve aforementioned problems, propose an image-level framework aims at adapting source-domain images with content-aligned source-target image...