A5055565056- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Image Retrieval and Classification Techniques
- Human Pose and Action Recognition
- Video Analysis and Summarization
- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Video Surveillance and Tracking Methods
- Gamma-ray bursts and supernovae
- Space Satellite Systems and Control
- Particle Detector Development and Performance
- Image and Object Detection Techniques
- X-ray Spectroscopy and Fluorescence Analysis
- Astro and Planetary Science
- 3D Surveying and Cultural Heritage
- Spacecraft Design and Technology
- Advanced Vision and Imaging
- Topic Modeling
- Gait Recognition and Analysis
- Satellite Communication Systems
- Atomic and Molecular Physics
- Text and Document Classification Technologies
- Face and Expression Recognition
- Natural Language Processing Techniques
- Cancer-related molecular mechanisms research
Institute of High Energy Physics
2022-2024
Chinese Academy of Sciences
2024
National Institute for Astrophysics
2020-2022
University of Udine
2022
Hebei Medical University
2022
Second Hospital of Hebei Medical University
2022
The University of Queensland
2014-2021
The University of Adelaide
2019-2020
Australian Centre for Robotic Vision
2019
Tongji University
2019
Person re-identification (Re-ID) has achieved great improvement with deep learning and a large amount of labelled training data. However, it remains challenging task for adapting model trained in source domain data to target only unlabelled available. In this work, we develop self-training method progressive augmentation framework (PAST) promote the performance progressively on dataset. Specially, our PAST consists two stages, namely, conservative stage promoting stage. The captures local...
The task in referring expression comprehension is to localize the object instance an image described by a phrased natural language. As language-to-vision matching task, key this problem learn discriminative feature that can adapt used. To avoid ambiguity, normally tends describe not only properties of referent itself, but also its relationships neighbourhood. capture and exploit important information we propose graph-based, language-guided attention mechanism. Being composed node component...
We propose an approach to estimate the 6DOF pose of a satellite, relative canonical pose, from single image. Such problem is crucial in many space proximity operations, such as docking, debris removal, and inter-spacecraft communications. Our combines machine learning geometric optimisation, by predicting coordinates set landmarks input image, associating their corresponding 3D points on priori reconstructed model, then solving for object using non-linear optimisation. not only novel this...
Deep networks excel in learning patterns from large amounts of data. On the other hand, many geometric vision tasks are specified as optimization problems. To seamlessly combine deep and vision, it is vital to perform end-to-end. Towards this aim, we present BPnP, a novel network module that backpropagates gradients through Perspective-n-Points (PnP) solver guide parameter updates neural network. Based on implicit differentiation, show ``self-contained" PnP can be derived accurately...
Vehicle instance retrieval (IR) often requires one to recognize the fine-grained visual differences between vehicles. Besides holistic appearance of vehicles which is easily affected by viewpoint variation and distortion, vehicle parts also provide crucial cues differentiate near-identical Motivated these observations, we introduce a <i>Part-Guided Attention Network</i> (PGAN) pinpoint prominent part regions effectively combine global local information for discriminative feature learning....
HERMES-TP/SP (High Energy Rapid Modular Ensemble of Satellites Technologic and Scientific Pathfinder) is a constellation six 3U nano-satellites hosting simple but innovative X-ray detectors, characterized by large energy band excellent temporal resolution, thus optimized for the monitoring Cosmic High transients such as Gamma Ray Bursts electromagnetic counterparts Gravitational Wave Events, determination their positions. The projects are funded Italian Ministry University Research Space...
Semantic information is important for video event detection. How to automatically discover, model, and utilize semantic facilitate detection has been a challenging problem. In this paper, we propose novel hierarchical which deliberately unifies the processes of underlying semantics discovery modeling from data. Specially, different most approaches based on manually pre-defined concepts, devise an effective model uncover by hierarchically capturing latent static-visual concepts in frame-level...
With the exponential increase of media data on web, fast retrieval is becoming a significant research topic in multimedia content analysis. Among variety techniques, learning binary embedding (hashing) functions one most popular approaches that can achieve scalable information large databases, and it mainly used near-duplicate search. However, till now hashing methods are specifically designed for at visual level rather than semantic level. In this paper, we propose state (VSBE) model to...
The enhanced x-ray timing and polarimetry mission (eXTP) is a flagship observatory for timing, spectroscopy developed by an international consortium. Thanks to its very large collecting area, good spectral resolution unprecedented capabilities, eXTP will explore the properties of matter propagation light in most extreme conditions found universe. will, addition, be powerful observatory. continuously monitor sky, enable multi-wavelength multi-messenger studies. currently phase B, which...
Nowadays the locations of social images play an important role in geographic knowledge discovery. However, most still lack location information, driving estimation for to have recently become active research topic. With rapid growth images, new challenges been posed: 1) data quality is issue because they are often associated with noises and error-prone user-generated content, such as junk comments misspelled words; 2) sparsity exists despite large volume, since them unevenly distributed...
Instance retrieval requires one to search for images that contain a particular object within large corpus. Recent studies show using image features generated by pooling convolutional layer feature maps (CFMs) of pretrained neural network (CNN) leads promising performance this task. However, due the global strategy adopted in those works, is less robust clutter and tends be contaminated irrelevant patterns. In article, we alleviate drawback proposing novel reranking algorithm CFMs refine...
The Large Area Detector (LAD) is the high-throughput, spectral-timing instrument onboard eXTP mission, a flagship mission of Chinese Academy Sciences and China National Space Administration, with large European participation coordinated by Italy Spain. currently performing its phase B study, target launch at end-2027. scientific payload includes four instruments (SFA, PFA, LAD WFM) offering unprecedented simultaneous wide-band X-ray timing polarimetry sensitivity. based on design originally...
Recently, neuron activations extracted from a pre-trained convolutional neural network (CNN) show promising performance in various visual tasks. However, due to the domain and task bias, using features generated model for image classification as representations instance retrieval is problematic. In this paper, we propose quartet-net learning improve discriminative power of CNN retrieval. The general idea map into space where similarity can be better evaluated. Our differs traditional...
Recently, hashing has been evidenced as an efficient and effective method to facilitate large-scale video retrieval. Most of existing methods are based on visual features, which expected capture the appearance videos. The intrinsic temporal pattern embedded in videos also shown its discriminative power for similarity search, is explored utilised some recent studies. However, how leverage strengths both aspects remains unknown.
Advances in machine learning have generated increasing enthusiasm for tasks that require high-level reasoning on top of perceptual capabilities, particularly over visual data. Such include, example, image captioning, question answering, and navigation. Their evaluation is however hindered by task-specific confounding factors dataset biases. In parallel, the existing benchmarks abstract are limited to synthetic stimuli (e.g. images simple shapes) do not capture challenges real-world We...
Existing video event classification approaches suffer from limited human-labeled semantic annotations. Weak annotations can be harvested Web-knowledge without involving any human interaction. However such weak are noisy, thus not effectively utilized distinguishing its reliability. In this paper, we propose a novel approach to automatically maximize the utility of (formalized as relevance shots target event) facilitate classification. A attention model is designed determine scores shots,...
Vehicle instance retrieval often requires one to recognize the fine-grained visual differences between vehicles. Besides holistic appearance of vehicles which is easily affected by viewpoint variation and distortion, vehicle parts also provide crucial cues differentiate near-identical Motivated these observations, we introduce a Part-Guided Attention Network (PGAN) pinpoint prominent part regions effectively combine global information for discriminative feature learning. PGAN first detects...
Long noncoding RNAs (lncRNAs) are correlated with cancer pathogenesis and prognosis. Many studies have shown that aberrant expression of MIR31HG is implicated in the progression patient However, biological function predictive value colorectal unclear.The correlation between clinicopathological characteristics patients was analyzed by collating information from The Cancer Genome Atlas (TCGA) database. Kaplan-Meier analysis, univariable multivariable Cox regression analysis were performed to...
Unsupervised cross-domain person re-identification (Re-ID) aims to adapt the information from labelled source domain an unlabelled target domain. Due lack of supervision in domain, it is crucial identify underlying similarity-and-dissimilarity relationships among samples In order use whole data efficiently mini-batch training, we apply a series memory modules maintain up-to-date representation entire dataset. Unlike simple exemplar previous works, propose novel multi-level network (MMN)...
The Bag-of-Words (BoW) models using the SIFT descriptors have achieved great success in content-based image retrieval over past decade. Recent studies show that neuron activations of convolutional neural networks (CNN) can be viewed as local descriptors, which aggregated into effective global for retrieval. However, little work has been done on these deep BoW models, especially case large visual vocabularies.