A5019281107- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- 3D Shape Modeling and Analysis
- Cutaneous Melanoma Detection and Management
- Remote Sensing and LiDAR Applications
- AI in cancer detection
- 3D Surveying and Cultural Heritage
- Time Series Analysis and Forecasting
- Medical Image Segmentation Techniques
- Sentiment Analysis and Opinion Mining
- Airway Management and Intubation Techniques
- Visual Attention and Saliency Detection
- Advanced Vision and Imaging
- Neural Networks and Applications
- Caching and Content Delivery
- Video Analysis and Summarization
- Simulation and Modeling Applications
- Subtitles and Audiovisual Media
- Computer Graphics and Visualization Techniques
- Software-Defined Networks and 5G
- Anomaly Detection Techniques and Applications
- Digital Imaging for Blood Diseases
- Nasal Surgery and Airway Studies
- Infrared Target Detection Methodologies
- Traffic Prediction and Management Techniques
Southwest Jiaotong University
2019-2024
Ministry of Transport
2024
Sichuan Agricultural University
2017-2019
Dense video captioning aims to detect and describe all events in untrimmed videos. This paper presents a dense network called Multi-Concept Cyclic Learning (MCCL), which to: (1) multiple concepts at the frame level leverage these provide temporal event cues; (2) establish cyclic co-learning between generator localizer within promote semantic perception localization. Specifically, weakly supervised concept detection is performed for each frame, detected embeddings are integrated into features...
Multimodal video sentiment analysis aims to integrate multiple modal information analyze the opinions and attitudes of speakers. Most previous work focuses on exploring semantic interactions intra- inter-modality. However, these works ignore reliability multimodality, i.e., modalities tend contain noise, ambiguity, missing modalities, etc. In addition, multimodal approaches treat different equally, largely ignoring their contributions. Furthermore, existing methods directly regress scores...
Point clouds data, as one kind of representation 3D objects, are the most primitive output obtained by sensors. Unlike 2D images, point disordered and unstructured. Hence it is not straightforward to apply classification techniques such convolution neural network analysis directly. To solve this problem, we propose a novel structure, named Attention-based Graph Convolution Networks (AGCN), extract features. Taking learning process message propagation between adjacent points, introduce an...
Dense video captioning aims to detect and describe all events in untrimmed videos. This paper presents a dense network called Multi-Concept Cyclic Learning (MCCL), which to: (1) multiple concepts at the frame level, using these enhance features provide temporal event cues; (2) design cyclic co-learning between generator localizer within promote semantic perception localization. Specifically, we perform weakly supervised concept detection for each frame, detected embeddings are integrated...
Unsupervised skin lesion segmentation offers several benefits, including conserving expert human resources, reducing discrepancies due to subjective labeling, and adapting novel envi- ronments. However, segmenting dermoscopic images without manual labeling guidance presents significant challenges image artifacts such as hair noise, blister subtle edge differences. To address these challenges, we introduce an innovative Uncertainty Self-Learning Network (USL-Net) designed for segmentation....
No-reference segmentation quality evaluation aims to evaluate the of image without any reference during application process. It usually depends on certain criteria describe a good with some prior knowledge. Therefore, there is need for precise description objects in and an integration representation In this paper, from perspective understanding semantic relationship between original results, we propose feature contrastive learning method. This method can enhance performance no-reference...
Unsupervised skin lesion segmentation offers several benefits, including conserving expert human resources, reducing discrepancies due to subjective labeling, and adapting novel environments. However, segmenting dermoscopic images without manual labeling guidance presents significant challenges image artifacts such as hair noise, blister subtle edge differences. To address these challenges, we introduce an innovative Uncertainty Self-Learning Network (USL-Net) designed for segmentation. The...
With the development of various 3D sensors, it has become easier for humans to obtain information in world, more and people turn their attention problem point clouds understanding. At present, most methods focus on directly extracting features from clouds, where feature extraction is performed by Multi-Layer Perception (MLP) fusion local pooling. However, they do not consider spatial relationship within sets. We propose a directed connected graph network (DCGN), which can effectively capture...