A5070443737- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Remote Sensing and LiDAR Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Optical Sensing Technologies
- Visual Attention and Saliency Detection
- Video Surveillance and Tracking Methods
- Optical measurement and interference techniques
- 3D Shape Modeling and Analysis
- 3D Surveying and Cultural Heritage
- Advanced Vision and Imaging
- Domain Adaptation and Few-Shot Learning
- Infrared Target Detection Methodologies
- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Veterinary Equine Medical Research
- Effects of Environmental Stressors on Livestock
- Computer Graphics and Visualization Techniques
- Image and Object Detection Techniques
- Animal Behavior and Welfare Studies
Linköping University
2016-2023
Video object segmentation (VOS) is a highly challenging problem since the initial mask, defining target object, only given at test-time. The main difficulty to effectively handle appearance changes and similar background objects, while maintaining accurate segmentation. Most previous approaches fine-tune networks on first frame, resulting in impractical frame-rates risk of overfitting. More recent methods integrate generative models, but either achieve limited robustness or require large...
The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by VOT initiative. Results of 71 trackers are presented; many state-of-the-art published at major computer vision conferences or in journals recent years. was composed four sub-challenges focusing on different tracking domains: (i) VOT-ST2021 focused short-term RGB, (ii) VOT-RT2021 "real-time" (iii) VOT-LT2021 long-term tracking, namely coping with target disappearance and reappearance...
Computer vision is a subcategory of artificial intelligence focused on extraction information from images and video. It provides compelling new means for objective orthopaedic gait assessment in horses using accessible hardware, such as smartphone, markerless motion analysis. This study aimed to explore the lameness capacity smartphone single camera (SC) computer application by comparing measurements vertical head pelvis an optical capture multi-camera (MC) system skin attached reflective...
Probabilistic methods for point set registration have demonstrated competitive results in recent years. These techniques estimate a probability distribution model of the clouds. While such representation has shown promise, it is highly sensitive to variations density 3D points. This fundamental problem primarily caused by changes sensor location across sets. We revisit foundations probabilistic paradigm. Contrary previous works, we underlying structure scene as latent distribution, and...
This letter introduces a framework for evaluation of the losses used in point set registration. In order loss to be useful with local optimizer, such as e.g. Levenberg-Marquardt, or expectation maximization (EM), it must monotonic respect sought transformation. motivates us introduce monotonicity violation probability (MVP) curves, and use these assess empirically many different distances, point-to-point, point-to-plane, plane-to-plane. We also shape-to-shape distance, based on Wasserstein...
Probabilistic methods for point set registration have interesting theoretical properties, such as linear complexity in the number of used points, and they easily generalize to joint multiple sets. In this work, we improve their recognition performance match state art. This is done by incorporating learned features, adding a von Mises-Fisher feature model each mixture component, using attention weights. We learn these jointly loss learning strategy (RLL) that directly uses error loss,...
Semantic segmentation of 3D point clouds is a challenging problem with numerous real-world applications. While deep learning has revolutionized the field image semantic segmentation, its impact on cloud data been limited so far. Recent attempts, based approaches (3D-CNNs), have achieved below-expected results. Such methods require voxelizations underlying data, leading to decreased spatial resolution and increased memory consumption. Additionally, 3D-CNNs greatly suffer from availability...
We address the highly challenging problem of video object segmentation. Given only initial mask, task is to segment target in subsequent frames. In order effectively handle appearance changes and similar background objects, a robust representation required. Previous approaches either rely on fine-tuning segmentation network first frame, or employ generative models. Although partially successful, these methods often suffer from impractically low frame rates unsatisfactory robustness. propose...
Video object segmentation (VOS) is a highly challenging problem since the initial mask, defining target object, only given at test-time. The main difficulty to effectively handle appearance changes and similar background objects, while maintaining accurate segmentation. Most previous approaches fine-tune networks on first frame, resulting in impractical frame-rates risk of overfitting. More recent methods integrate generative models, but either achieve limited robustness or require large...
In this paper we introduce an efficient method to unwrap multi-frequency phase estimates for time-of-flight ranging. The algorithm generates multiple depth hypotheses and uses a spatial kernel density estimate (KDE) rank them. confidence produced by the KDE is also effective means detect outliers. We new closed-form expression noise prediction, that better fits real data. applied decoding Kinect v2 sensor, compared Microsoft SDK open source driver libfreenect2. intended use case scenes with...
Probabilistic methods for point set registration have demonstrated competitive results in recent years. These techniques estimate a probability distribution model of the clouds. While such representation has shown promise, it is highly sensitive to variations density 3D points. This fundamental problem primarily caused by changes sensor location across sets. We revisit foundations probabilistic paradigm. Contrary previous works, we underlying structure scene as latent distribution, and...
Probabilistic methods for point set registration have interesting theoretical properties, such as linear complexity in the number of used points, and they easily generalize to joint multiple sets. In this work, we improve their recognition performance match state art. This is done by incorporating learned features, adding a von Mises-Fisher feature model each mixture component, using attention weights. We learn these jointly loss learning strategy (RLL) that directly uses error loss,...