A5039567778- Human Pose and Action Recognition
- Anomaly Detection Techniques and Applications
- Multimodal Machine Learning Applications
- Robot Manipulation and Learning
- Gait Recognition and Analysis
- Video Surveillance and Tracking Methods
- Landslides and related hazards
- Industrial Vision Systems and Defect Detection
- Diabetic Foot Ulcer Assessment and Management
- Advanced Image Processing Techniques
- Climate variability and models
- Speech and dialogue systems
- Plant Water Relations and Carbon Dynamics
- Adversarial Robustness in Machine Learning
- Physical Activity and Education Research
- Soil and Land Suitability Analysis
- Advanced Neural Network Applications
- Image and Signal Denoising Methods
- Flood Risk Assessment and Management
- Tree Root and Stability Studies
- Sparse and Compressive Sensing Techniques
- Environmental and Agricultural Sciences
Beijing University of Technology
2022-2025
As a powerful statistical signal modeling technique, sparse representation has been widely used in various image restoration (IR) applications. The sparsity-based methods have achieved leading performance the past few decades. However, recent years it surpassed by other methods, especially deep learning based methods. In this paper, we address question that whether can be competitive again. way answer is to redesign with architecture. To specific, propose an end-to-end architecture follows...
3D human pose estimation (3DHPE) in images aims at estimating joint positions from images. The existing 3DHPE methods usually define the loss function as error measured by Euclidean distance between locations of predicted joints and ground truth joints, which confuses two different kinds errors with obviously characteristics should not be processed equally: caused structures others. However, representations are suitable to distinguish these errors. In order tackle this problem, we propose a...
Affordance refers to the interactable functional properties of an object, and affordance segmentation aims pixel-level segment object parts in a given image, which is crucial for various interactive vision tasks. Existing methods address problem by utilizing only image features, they can hardly solve problems interference between adjacent pixels complex scenes, inability generalize open-world. To tackle these problems, we propose novel open-vocabulary task benchmark dataset, approach with...
Visual affordance recognition is an important research topic in robotics, human-computer interaction, and other computer vision tasks. In recent years, deep learning-based methods have achieved remarkable performance. However, there no unified intensive survey of these up to now. Therefore, this article reviews investigates existing from a comprehensive perspective, hoping pursue greater acceleration domain. Specifically, first classifies into five tasks, delves the methodologies each task,...
The focus of this paper is on the grassland productivity response to drought under background climate change. There an established lag impact ecosystems events, which may have additional effects subsequent events. Meanwhile, due change interference, influence over past 50 years not simply equal algebraic sum all historical In Inner Mongolia grassland, precipitation deficit plays a leading role in causing drought. Therefore, taking into consideration impacts effect and change, paper, we net...
3D human pose estimation (3DHPE) in images aims at estimating joint positions from images.The state-of-theart for 3DHPE is dominated by deep learning model whose accuracy obviously affected loss functions.The existing methods usually define the function as error measured Euclidean distance between locations of predicted joints and ground truth joints, which confuses two different kinds errors: caused structures others.But fact, characteristics these errors are should not be processed...
<title>Abstract</title> Thanks to the development of 2D keypoint detectors, monocular 3D human pose estimation (HPE) via 2D-to-3D lifting approaches have achieved remarkable improvements. However, HPE is still a challenging problem due inherent depth ambiguities and occlusions. Recently, diffusion models great success in field image generation. Inspired by this, we transform into reverse process, propose dual-branch model that could fully explore global local correlations between joints....
Traditional affordance learning tasks aim to understand object's interactive functions in an image, such as recognition and detection. However, these cannot determine whether the object is currently interacting, which crucial for many follow-up tasks, including robotic manipulation planning task. To fill this gap, paper proposes a novel affrodance state (OAS) task, i.e., simultaneously recognizing affordances partner objects that are interacting with it. Accordingly, facilitate application...
Slope collapse is one of the most severe natural disaster threats, and accurately predicting slope deformation important to avoid occurrence disaster. However, single prediction model has some problems, such as poor stability, lower accuracy data fluctuation. Obviously, it necessary establish a combination predict deformation. Here, we used GFW-Fisher optimal segmentation method multi-scale model. Our results indicated that determination coefficient linear model, weighted geometric average...
Most slope collapse accidents are indicated by certain signs before their occurrence, and unnecessary losses can be avoided predicting deformation. However, the early warning of deformation often misjudged. It is necessary to establish a method determine appropriate in sliding thresholds. Here, better understand impact different scales on thresholds, we used Fisher optimal segmentation threshold model based speed acceleration at spatial scales. Our results that accuracy thresholds surface...
Object-object affordance recognition aims to recognize the interactive relations between an object and other objects, which plays a crucial role in task decision-making selection of industrial robots. To address problem interference from complex interaction relations, we propose object-object affordances via relational phrase learning. The phrases are used as knowledge prior improve expression. In addition, multi-scale feature pooling aggregation module enhance visual representation images....