A5101526745- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Image Retrieval and Classification Techniques
- Advanced Data Storage Technologies
- Advanced Neural Network Applications
- Distributed and Parallel Computing Systems
- MicroRNA in disease regulation
- Video Analysis and Summarization
- Cancer-related molecular mechanisms research
- Advanced Vision and Imaging
- Neural Networks and Applications
- Molecular Biology Techniques and Applications
- Robotics and Sensor-Based Localization
- Video Surveillance and Tracking Methods
- Cloud Computing and Remote Desktop Technologies
- Lipid metabolism and disorders
- Text and Document Classification Technologies
- Genomic variations and chromosomal abnormalities
- Educational Robotics and Engineering
- Ferroelectric and Piezoelectric Materials
- Forensic and Genetic Research
- Domain Adaptation and Few-Shot Learning
- Prenatal Screening and Diagnostics
- Advanced Malware Detection Techniques
- Organic Light-Emitting Diodes Research
University of Hong Kong
2025
Beijing National Laboratory for Molecular Sciences
2025
Peking University
2025
State Key Laboratory of Rare Earth Materials Chemistry and Application
2025
Sichuan University
2021-2023
West China Second University Hospital of Sichuan University
2022-2023
Guilin University of Electronic Technology
2022
Baidu (China)
2021-2022
Bellevue Hospital Center
2021-2022
University of Electronic Science and Technology of China
2014-2021
Recently, MLP-based vision backbones emerge. architectures with less inductive bias achieve competitive performance in image recognition compared CNNs and Transformers. Among them, spatial-shift MLP (S$^2$-MLP), adopting the straightforward operation, achieves better than pioneering works including MLP-mixer ResMLP. More recently, using smaller patches a pyramid structure, Vision Permutator (ViP) Global Filter Network (GFNet) S$^2$-MLP. In this paper, we improve S$^2$-MLP backbone. We expand...
Colloidal quantum dots (QDs) have illuminated computer monitors and television screens due to their fascinating color-tunable properties depending on the size. Here, electroluminescence (EL) wavelength of perovskite LEDs was tuned via atomic layer number (ALN) nanoplates (NPs) instead "size" in conventional QDs. We demonstrated efficient with controllably tailored emission from n = 3, 4, 5, ≥7 ALN NPs specific discrete major peaks at 607, 638, 669, 728 nanometers. These peak external...
Most of current visual search systems focus on image-to-image (point-to-point) such as image and object retrieval. Nevertheless, fast image-to-video (point-to-set) is much less exploited. This paper tackles instance in videos, where efficient point-to-set matching essential. Through jointly optimizing vector quantization hashing, we propose compressive method to compressM proposals extracted from each video into only k binary codes, ≪ M. Then the similarity between query whole can be...
By exploiting the cross-modal attention, cross-BERT methods have achieved state-of-the-art accuracy in retrieval. Nevertheless, heavy text-image interactions model are prohibitively slow for large-scale Late-interaction trade off retrieval and efficiency by interaction only late stage, attaining a satisfactory speed. In this work, we propose an inflating shrinking approach to further boost of late-interaction methods. The operation plugs several codes input encoder exploit more thoroughly...
Bilinear pooling has achieved an excellent performance in many computer vision tasks such as fine-grained classification, scene recognition and texture recognition. However, the high-dimension features from bilinear can sometimes be inefficient prone to over-fitting. Random Maclaurin (RM) is a widely used GPU-friendly approximation method reduce dimensionality of features. achieve good performance, large projection matrices are usually required practice, making it costly computation memory....
Abstract Background Patients with biallelic variants in the lanosterol synthase ( LSS ) gene has been reported to exhibit phenotypes as follows: non‐syndromic form of hypotrichosis, congenital cataracts, and alopecia intellectual disability or growth retardation. However, genotype–phenotype correlations are still not completely clear. Methods In this study, we a Chinese girl who had cataracts hypotrichosis. The trio exome sequencing was performed elucidate genetic cause patient. Results We...
This paper tackles the problem of efficient and effective object instance search in videos. To effectively capture relevance between a query video frames precisely localize particular object, we leverage proposals to improve quality However, hundreds obtained from each frame could result unaffordable memory computational cost. this end, present simple yet hierarchical prototype encoding (HOPE) model accelerate without sacrificing accuracy, which exploits both spatial temporal self-similarity...
Since an image can be perceived by customers in few seconds, it is effective medium for advertising and adored advertisers. Baidu, as one of the lead search companies world, receives billions text queries per day. How to feed attractive images capture customers' attentions core task Baidu advertising. Traditionally, query-to-image tackled matching query with title. Nevertheless, title-based relies on high-quality titles, which are not easy obtained or unavailable some cases. A more reliable...
Video advertisements may grasp customers' attention instantly and are often adored by advertisers. Since the corpus is vast, achieving an efficient query-to-video search can be challenging. Because traditional approximate nearest neighborhood (ANN) methods based simple similarities (e.g., cosine or inner products) on embedding vectors. They not sufficient for bridging modal gap between a text query video typically only achieve sub-optimal performance in search. Tree-based deep model (TDM)...
In order to make more accurate and scientific predictions of the industrial economy, we integrated method BP artificial neural network with grey relational analysis together as an Industrial economic forecasting model in this study, applied it predict gross output Huainan City China. This paper examines two most recent patents between May 2012 January 2013 area forecasting. The efficiency effectiveness new is tested by comparing predicted results stepwise regression GM (1, 1) models....
Due to their low storage capacity and fast retrieval speed, hashing techniques have received much attention in cross-modal retrieval. However, there are some issues that need be further explored. First, existing methods use the labels construct semantic similarity matrix between pairwise data, ignoring potential manifold structure heterogeneous data. Second, underestimate importance of multi-label gaps different class labels, making learned hash codes less discriminative. Third, few them...
Automatically discovering common visual patterns from images and videos is a useful but challenging task. On the one hand, definition of rather ambiguous, it refers to spatial composition frequently occurring primitives which correspond local features, semantic parts or objects. For example, wheels body car could be seen as different primitives, while whole can also an individual primitive. other there exhibit large variations in appearance structures even within same kind pattern, makes...
In this paper, we implement a real-time method for incrementally stitching aerial images over large area. To collect high-quality images, the camera is mounted on gimbal. Being isolated from vibration of drone, can provide relatively stable and less blurry images. achieve fast mapping, instead using traditional feature extraction matching steps, algorithm only relies position orientation blending. The mosaic fused updated based adaptive weight multi-band blending algorithm. matrix Laplacian...
The discriminant cut algorithm is used to detect coastlines in synthetic aperture radar (SAR) images. proposed approach a region-based one, which able capture and utilize spatial information the image. real SAR images, e.g. ALOS-1 PALSAR COSMO-SkyMed together with in-situ GPS data were collected validate performance of for coastline detection accuracy better than 4 times image resolution. efficiency also tested.
Type III Bartter syndrome (BS), often known as classic is caused by variants in CLCNKB gene, which encoding the basolateral chloride channel protein ClC-Kb, and characterized renal salt wasting, hypokalemia, metabolic alkalosis, increased renin, aldosterone levels.A 2-year-old boy presented severe malnutrition, alkalosis hypokalemia was clinically diagnosed with BS. The trio exome sequencing (ES) performed to discover genetic cause of this patient, followed validation using Sanger...
Abstract Background Short‐rib thoracic dysplasia (SRTD) and Joubert syndrome (JS) are rare genetic ciliopathies, individuals with either can manifest cerebellar malformation variable developmental delays. However, neither of these conditions is easily diagnosed during pregnancy due to a limited fetal phenotype. Here, we investigated fetus that was initially observed have short limbs polydactyly discovered compound heterozygous pathogenesis through exome sequencing (ES). Methods Simultaneous...
In recent years, China has developed modern agriculture energetically. An effective information framework is an important way to provide farms with agricultural services and improve farmer's production technology their income. The mountain areas in central are dominated by agriculture, such as Jiangxi province, Anhui province. These area exist many problems on striking regional difference, unbalanced economic development, low cultural level of most the farmers requirement national policy,...