A5020970600- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Domain Adaptation and Few-Shot Learning
- Machine Learning and Data Classification
- Remote Sensing and LiDAR Applications
- Image Retrieval and Classification Techniques
- Visual Attention and Saliency Detection
- Medical Image Segmentation Techniques
- Infrared Target Detection Methodologies
- Automated Road and Building Extraction
- Soybean genetics and cultivation
- Robotics and Sensor-Based Localization
- Advanced Image Processing Techniques
- Remote-Sensing Image Classification
- Plant Molecular Biology Research
- High-Velocity Impact and Material Behavior
- Fire effects on concrete materials
- Structural Response to Dynamic Loads
- Multimodal Machine Learning Applications
- Image Processing Techniques and Applications
- Seed Germination and Physiology
- AI in cancer detection
- Advanced Vision and Imaging
Changsha Mining and Metallurgy Research Institute (China)
2024
Horizon Robotics (China)
2019-2021
Huazhong University of Science and Technology
2019
Institute of Automation
2015-2017
Chinese Academy of Sciences
2014-2017
Letting a deep network be aware of the quality its own predictions is an interesting yet important problem. In task instance segmentation, confidence classification used as mask score in most segmentation frameworks. However, quality, quantified IoU between and ground truth, usually not well correlated with score. this paper, we study problem propose Mask Scoring R-CNN which contains block to learn predicted masks. The proposed takes feature corresponding together regress IoU. scoring...
Object detection and instance segmentation are two fundamental computer vision tasks. They closely correlated but their relations have not yet been fully explored in most previous work. This paper presents RDSNet, a novel deep architecture for reciprocal object segmentation. To reciprocate these tasks, we design two-stream structure to learn features on both the level (i.e., bounding boxes) pixel masks) jointly. Within this structure, information from streams is fused alternately, namely...
In this paper, we propose a graph cuts (GC) based probability propagation approach to automatically extract road network from complex remote sensing images. First, the support vector machine (SVM) classifier with sigmoid model is applied assign each pixel posterior of being labelled as class, which avoids weaknesses hard labels in general SVM. Then GC algorithm employed keep extracted results smooth and coherent, can reduce connections between roads road-like objects. Finally,...
Object detection and instance segmentation are two fundamental computer vision tasks. They closely correlated but their relations have not yet been fully explored in most previous work. This paper presents RDSNet, a novel deep architecture for reciprocal object segmentation. To reciprocate these tasks, we design two-stream structure to learn features on both the level (i.e., bounding boxes) pixel masks) jointly. Within this structure, information from streams is fused alternately, namely...
In this paper, we propose a novel motion-guided attention module to implant the spatial and time consistency in correlation map of current frame with historical frames. Unlike other mask propagation based methods, our method regards previous as strong prior instead concatenating it or feature for propagation. Additionally, reduce gap between training testing phase, an improved optimization strategy, named sequence learning, which feeds video chronological order into end-to-end network...
Deep convolutional neural networks (CNNs) have shown outstanding performance in object recognition from natural images. In contrast, remote sensing images is more challenging, due to the complex background and inadequate data for training a deep network with huge number of parameters. We propose unified CNN, called DeepPlane, simultaneously detect position classify category aircraft This model consists two correlative networks: first one designed generate proposals as well feature maps...
Video object segmentation (VOS) aims at pixel-level tracking given only the annotations in first frame. Due to large visual variations of objects video and lack training samples, it remains a difficult task despite upsurging development deep learning. Toward solving VOS problem, we bring several new insights by proposed unified framework consisting proposal, components. The proposal network transfers objectness information as generic knowledge into VOS; identifies target from proposals; is...
Fine-structured object segmentation is a challenging problem in community. There are mainly two difficulties that can seriously degrade the quality: 1) insufficient interactions on fine structures due to high demand of time and manual efforts, 2) shrinking bias discourages long boundaries. To address these issues, we develop novel method within graph cut framework. First, commonly used operation scribbling or dragging bounding boxes replaced by loosely drawing few rectangles, thus...
Letting a deep network be aware of the quality its own predictions is an interesting yet important problem. In task instance segmentation, confidence classification used as mask score in most segmentation frameworks. However, quality, quantified IoU between and ground truth, usually not well correlated with score. this paper, we study problem propose Mask Scoring R-CNN which contains block to learn predicted masks. The proposed takes feature corresponding together regress IoU. scoring...
During the life cycle of a plant, seed germination is crucial. Upon ingestion water, dry seeds resumed energy metabolism and cellular repair. To dissect complex mechanisms at very beginning germination, two approaches including transcriptome small RNA sequencing were conducted during water imbibition process mung bean compared with seed. The analysis identified 10,108 differentially expressed genes (DEGs) between imbibed seeds. KEGG enrichment analyses demonstrated numerous DEGs involved in...
In this paper, we present a novel method for the challenging task of fine-structured (FS) object segmentation. This is formulated as label propagation problem on an affinity graph. To enhance completeness and connectivity FS objects, introduce neighborhood system combining both local nonlocal connections, together with robust scheme edge weight calculation. Additionally, region cost incorporated into energy function to further maintain fine parts where hard reach. An appealing advantage...
Random walks (RW) is a popular technique for object segmentation. Apart from the satisfactory performance in various applications, its most appealing advantage computational efficiency. However, RW often fails to produce complete and connected results fine-structured (FS) To utilize high efficiency overcome drawbacks tackling FS objects, we develop novel approach within framework. Specifically, propose introduce labeling preference learned image data into model guide propagation of random...