A5102002226- Advanced Neural Network Applications
- Autonomous Vehicle Technology and Safety
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Traffic and Road Safety
- Cell Image Analysis Techniques
- Marine and coastal ecosystems
- Image Processing Techniques and Applications
- Anomaly Detection Techniques and Applications
- Image Processing and 3D Reconstruction
- Aquatic Ecosystems and Phytoplankton Dynamics
- Software Reliability and Analysis Research
- Robotics and Sensor-Based Localization
- Software Engineering Research
- Human Pose and Action Recognition
- Time Series Analysis and Forecasting
- Software System Performance and Reliability
- Multimodal Machine Learning Applications
Peking University
2024
Tongji University
2023
Hubei University
2023
The multi-modality and stochastic characteristics of human behavior make motion prediction a highly challenging task, which is critical for autonomous driving. While deep learning approaches have demonstrated their great potential in this area, it still remains unsolved to establish connection between multiple driving scenes (e.g., merging, roundabout, intersection) the design models. Current learning-based methods typically use one unified model predict trajectories different scenarios, may...
Driving scene understanding is to obtain compre-hensive information through the sensor data and provide a basis for downstream tasks, which indispensable safety of self-driving vehicles. Specific perception such as object detection graph generation, are commonly used. However, results these tasks only equivalent characterization sampling from high-dimensional features, not sufficient represent scenario. In addition, goal inconsistent with human driving that just focuses on what may affect...
Self-supervised multi-frame methods have currently achieved promising results in depth estimation. However, these often suffer from mismatch problems due to the moving objects, which break static assumption. Additionally, unfairness can occur when calculating photometric errors high-freq or low-texture regions of images. To address issues, existing approaches use additional semantic priori black-box networks separate objects and improve model only at loss level. Therefore, we propose...
Driving scene understanding is to obtain comprehensive information through the sensor data and provide a basis for downstream tasks, which indispensable safety of self-driving vehicles. Specific perception such as object detection graph generation, are commonly used. However, results these tasks only equivalent characterization sampling from high-dimensional features, not sufficient represent scenario. In addition, goal inconsistent with human driving that just focuses on what may affect...
Graph neural network is an effective deep learning framework for graph data.Existing research has introduced different variants of networks into the field software defects and achieved promising results.However, model based on previous essentially transductive, applied to a single fixed graph, often ignores direction weight edges when modeling network.In practice, systems are dynamically evolving.Furthermore, in modeling, factors that worth considering.Based inductive network, we proposed...
One essential task for autonomous driving is to accurately predict the future motions of surrounding traffic agents. Recently, Graph Neural Network (GNN) approaches have shown potential in motion prediction due fact that information scenario can be inherently formed into a graph structure. However, existing are limited using static representing states current timestamp, ignoring scenario's derivation. In this work, we propose Future (FGNet) two-stage GNN-based model accurate and real-time...