A5070492669- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Advanced Vision and Imaging
- Genetic Mapping and Diversity in Plants and Animals
- Robotics and Sensor-Based Localization
- Multimodal Machine Learning Applications
- 3D Shape Modeling and Analysis
- Computer Graphics and Visualization Techniques
- 3D Surveying and Cultural Heritage
- Pancreatic function and diabetes
- Domain Adaptation and Few-Shot Learning
- Human Pose and Action Recognition
- Visual Attention and Saliency Detection
- Genetics and Plant Breeding
- Generative Adversarial Networks and Image Synthesis
- Video Analysis and Summarization
- Wheat and Barley Genetics and Pathology
- RNA modifications and cancer
- Face recognition and analysis
- Crop Yield and Soil Fertility
- Remote Sensing and LiDAR Applications
- Video Surveillance and Tracking Methods
- Speech and Audio Processing
- Smart Agriculture and AI
- Genetic and phenotypic traits in livestock
Microsoft Research Asia (China)
2018-2024
National Engineering Research Center for Information Technology in Agriculture
2017-2024
First Affiliated Hospital of Zhengzhou University
2014-2024
Jingchu University of Technology
2022-2023
University of Miami
2023
Huazhong Agricultural University
2021-2023
Microsoft Research (United Kingdom)
2018-2023
Ningbo First Hospital
2016-2023
Northwest A&F University
2019-2022
Shanxi Agricultural University
2022
Localizing objects in the real 3D space, which plays a crucial role scene understanding, is particularly challenging given only single RGB image due to geometric information loss during imagery projection. We propose MonoGRNet for amodal object localization from monocular via reasoning both observed 2D projection and unobserved depth dimension. single, unified network composed of four task-specific subnetworks, responsible detection, instance estimation (IDE), local corner regression. Unlike...
With the rapid development of genetic analysis techniques and crop population size, phenotyping has become bottleneck restricting breeding. Breaking through this will require phenomics, defined as accurate, high-throughput acquisition multi-dimensional phenotypes during growth at organism-wide levels, ranging from cells to organs, individual plants, plots, fields. Here we offer an overview phenomics research technological platform viewpoints various scales, including microscopic,...
In this paper, we study the problem of 3D object detection from stereo images, in which key challenge is how to effectively utilize information. Different previous methods using pixel-level depth maps, propose employ anchors explicitly construct object-level correspondences between regions interest deep neural network learns detect and triangulate targeted space. We also introduce a cost-efficient channel reweighting strategy that enhances representational features weakens noisy signals...
In this paper, we present a joint multi-task learning framework for semantic segmentation and boundary detection. The critical component in the is iterative pyramid context module (PCM), which couples two tasks stores shared latent semantics to interact between tasks. For detection, propose novel spatial gradient fusion suppress non-semantic edges. As detection dual task of segmentation, introduce loss function with consistency constraint improve pixel accuracy segmentation. Our extensive...
Weakly-supervised temporal action localization aims to learn detecting intervals of classes with only video-level labels. To this end, it is crucial separate frames from the background (i.e., not belonging any classes). In paper, we present a new perspective on where they are modeled as out-of-distribution samples regarding their inconsistency. Then, can be detected by estimating probability each frame being out-of-distribution, known uncertainty, but infeasible directly uncertainty without...
A crucial task in scene understanding is 3D object detection, which aims to detect and localize the bounding boxes of objects belonging specific classes. Existing detectors heavily rely on annotated during training, while these annotations could be expensive obtain only accessible limited scenarios. Weakly supervised learning a promising approach reducing annotation requirement, but existing weakly are mostly for 2D detection rather than 3D. In this work, we propose VS3D, framework from...
Error propagation is a general but crucial problem in online semi-supervised video object segmentation. We aim to suppress error through correction mechanism with high reliability. The key insight disentangle the from conventional mask process reliable cues. introduce two modulators, and separately perform channel-wise recalibration on target frame embeddings according local temporal correlations references respectively. Specifically, we assemble modulators cascaded propagation-correction...
Despite surgical and chemotherapeutic advances over the past few decades, prognosis for ovarian cancer remains very poor. Although cyclin-dependent kinase (CDK) 9 has an established pathogenic role in various cancers, its function poorly defined. The purpose of this study was to evaluate expression CDK9 therapeutic potential cancer. determined by immunohistochemistry a unique tissue microarray constructed with paired primary, metastatic, recurrent tumor tissues from 26 patients. highly...
Stalk lodging is an impediment to improving profitability and production efficiency in maize. Lodging resistance, a comprehensive indicator appraise genotypes, requires both characterization of mechanical properties laboratory investigation percentage field. However, situ maize resistance still remains poor. The aim this study was develop indicator, named cumulative index (CLI), based on percentages at different wind speeds for evaluating cultivars, evaluate the accuracy reliability...
Summary High‐throughput phenotyping is increasingly becoming an important tool for rapid advancement of genetic gain in breeding programmes. Manual vascular bundles tedious and time‐consuming, which lags behind the development functional genomics maize. More robust automated techniques traits at high‐throughput are urgently needed large crop populations. In this study, we developed a standard process stem micro‐CT data acquisition automatic CT image pipeline to obtain bundle stems including...
Previous works on video object segmentation (VOS) are trained densely annotated videos. Nevertheless, acquiring annotations in pixel level is expensive and time-consuming. In this work, we demonstrate the feasibility of training a satisfactory VOS model sparsely videos—we merely require two labeled frames per while performance sustained. We term novel paradigm as two-shot segmentation, or for short. The underlying idea to generate pseudo labels unlabeled during optimize combination...
Referring Video Object Segmentation (R-VOS) is a challenging task that aims to segment an object in video based on linguistic expression. Most existing R-VOS methods have critical assumption: the referred must appear video. This assumption, which we refer as "semantic consensus", often violated real-world scenarios, where expression may be queried against false videos. In this work, highlight need for robust model can handle semantic mismatches. Accordingly, propose extended called Robust (R...
Large Reconstruction Models (LRMs) have recently become a popular method for creating 3D foundational models. Training reconstruction models with 2D visual data traditionally requires prior knowledge of camera poses the training samples, process that is both time-consuming and prone to errors. Consequently, has been confined either synthetic datasets or small-scale annotated poses. In this study, we investigate feasibility using unposed video various objects. We introduce UVRM, novel model...
The power of modern image matching approaches is still fundamentally limited by the abrupt scale changes in images. In this paper, we propose a scale-invariant approach to tackling very large variation views. Drawing inspiration from space theory, start with encoding image's into compact multi-scale representation. Then, rather than trying find exact feature matches all one step, progressive two-stage approach. First, determine related levels space, enclosing inlier correspondences, based on...
In this paper, we tackle the accurate and consistent Structure from Motion (SfM) problem, in particular camera registration, far exceeding memory of a single computer parallel. Different previous methods which drastically simplify parameters SfM sacrifice accuracy final reconstruction, try to preserve connectivities among cameras by proposing clustering algorithm divide large problem into smaller sub-problems terms clusters with overlapping. We then exploit hybrid formulation that applies...
Semantic segmentation is pixel-wise classification which retains critical spatial information. The “feature map reuse” has been commonly adopted in CNN based approaches to take advantage of feature maps the early layers for later reconstruction. Along this direction, we go a step further by proposing fully dense neural network with an encoderdecoder structure that abbreviate as FDNet. For each stage decoder module, all previous blocks are adaptively aggregated feedforward input. On one hand,...
Detecting and localizing objects in the real 3D space, which plays a crucial role scene understanding, is particularly challenging given only monocular image due to geometric information loss during imagery projection. We propose MonoGRNet for amodal object detection from via reasoning both observed 2D projection unobserved depth dimension. decomposes task into four sub-tasks including detection, instance-level estimation, projected center estimation local corner regression. The...