A5027159188- Advanced Vision and Imaging
- Computer Graphics and Visualization Techniques
- Image Enhancement Techniques
- Optical measurement and interference techniques
- 3D Shape Modeling and Analysis
- Advanced Image Processing Techniques
- 3D Surveying and Cultural Heritage
- Image Processing and 3D Reconstruction
- Anomaly Detection Techniques and Applications
- Advanced Numerical Analysis Techniques
- Advanced Fiber Optic Sensors
- Fire Detection and Safety Systems
- Robotics and Sensor-Based Localization
- Handwritten Text Recognition Techniques
- Color perception and design
- Advanced X-ray Imaging Techniques
- Human Pose and Action Recognition
- Semiconductor Lasers and Optical Devices
- Photonic and Optical Devices
- Morphological variations and asymmetry
- Advanced Neural Network Applications
- Optical Polarization and Ellipsometry
- Neural Networks and Applications
- Color Science and Applications
Horizon Robotics (China)
2025
Nankai University
2021-2024
University of Electronic Science and Technology of China
2019-2021
In this paper, we propose a deep learning architecture that produces accurate dense depth for the outdoor scene from single color image and sparse depth. Inspired by indoor completion, our network estimates surface normals as intermediate representation to produce depth, can be trained end-to-end. With modified encoder-decoder structure, effectively fuses LiDAR To address specific challenges, predicts confidence mask handle mixed signals near foreground boundaries due occlusion, combines...
The channel redundancy in feature maps of convolutional neural networks (CNNs) results the large consumption memories and computational resources. In this work, we design a novel Slim Convolution (SlimConv) module to boost performance CNNs by reducing redundancies. Our SlimConv consists three main steps: Reconstruct, Transform Fuse, through which features are splitted reorganized more efficient way, such that learned weights can be compressed effectively. particular, core our model is weight...
Image relighting is attracting increasing interest due to its various applications. From a research perspective, im-age can be exploited conduct both image normalization for domain adaptation, and also data augmentation. It has multiple direct uses photo montage aesthetic enhancement. In this paper, we review the NTIRE 2021 depth guided challenge.We rely on VIDIT dataset each of our two challenge tracks, including information. The first track one-to-one where goal transform illumination...
Existing Neural Radiance Fields (NeRF) methods suffer from the existence of reflective objects, often resulting in blurry or distorted rendering. Instead calculating a single radiance field, we propose multi-space neural field (MS-NeRF) that represents scene using group feature fields parallel sub-spaces, which leads to better understanding network toward and refractive objects. Our scheme works as an enhancement existing NeRF methods, with only small computational overheads needed for...
Neural implicit methods have achieved high-quality 3D object surfaces under slight specular highlights. However, high reflections (HSR) often appear in front of target objects when we capture them through glasses. The complex ambiguity these scenes violates the multi-view consistency, then makes it challenging for recent to reconstruct correctly. To remedy this issue, present a novel surface reconstruction framework, NeuS-HSR, based on neural rendering. In NeuSHSR, is parameterized as an...
Existing Neural Radiance Fields (NeRF) methods suffer from the existence of reflective objects, often resulting in blurry or distorted rendering. Instead calculating a single radiance field, we propose multi-space neural field (MS-NeRF) that represents scene using group feature fields parallel sub-spaces, which leads to better understanding network toward and refractive objects. Our scheme works as an enhancement existing NeRF methods, with only small computational overheads needed for...
Recently RGB-D sensors have become very popular in the area of Simultaneous Localisation and Mapping (SLAM). The SLAM approach relies heavily on accuracy input depth map. However, refraction reflection transparent objects will result false cameras, which makes traditional algorithm unable to work correctly presence objects. In this paper, we propose a novel called transfusion that allows object existence recovery video input. Our method is composed two parts. Transparent Objects Cut...
In this paper, we propose a deep learning architecture that produces accurate dense depth for the outdoor scene from single color image and sparse depth. Inspired by indoor completion, our network estimates surface normals as intermediate representation to produce depth, can be trained end-to-end. With modified encoder-decoder structure, effectively fuses LiDAR To address specific challenges, predicts confidence mask handle mixed signals near foreground boundaries due occlusion, combines...
Learning from multi-view images using neural implicit signed distance functions shows impressive performance on 3D Reconstruction of opaque objects. However, existing methods struggle to reconstruct accurate geometry when applied translucent objects due the non-negligible bias in their rendering function. To address inaccuracies model, we have reparameterized density function radiance field by incorporating an estimated constant extinction coefficient. This modification forms basis our...
Novel view synthesis of smoke scenes presents a challenging problem. Previous neural approaches have suffered from inadequate quality and inefficient training. We introduce NeuSmoke, an efficient framework for dynamic reconstruction using transportation fields, enabling high-quality density novel-view multi-view videos. Our consists two stages. In the first stage, we design novel fluid field representation, integrating transport equation with fields. This includes adaptive embedding multiple...
Recently, 3D Gaussian Splatting (3DGS) has achieved significant performance on indoor surface reconstruction and open-vocabulary segmentation. This paper presents GLS, a unified framework of segmentation based 3DGS. GLS extends two fields by exploring the correlation between them. For reconstruction, we introduce normal prior as geometric cue to guide rendered normal, use error optimize depth. segmentation, employ 2D CLIP features instance utilize DEVA masks enhance their view consistency....
3D scene reconstruction is a foundational problem in computer vision. Despite recent advancements Neural Implicit Representations (NIR), existing methods often lack editability and compositional flexibility, limiting their use scenarios requiring high interactivity object-level manipulation. In this paper, we introduce the Gaussian Object Carver (GOC), novel, efficient, scalable framework for object-compositional reconstruction. GOC leverages Splatting (GS), enriched with monocular geometry...
Caustics are challenging light transport effects for photo-realistic rendering. Photon mapping techniques play a fundamental role in rendering caustics. However, photon methods render single caustics under the stationary source fixed scene view. They require significant storage and computing resources to produce high-quality results. In this paper, we propose efficiently more diverse of with camera moving. We present novel learning-based volume approach implicit representations our proposed...
Outdoor vision robotic systems and autonomous cars suffer from many image-quality issues, particularly haze, defocus blur, motion which we will define generically as "blindness issues". These blindness issues may seriously affect the performance of could lead to unsafe decisions being made. However, existing solutions either focus on one type only or lack ability estimate degree accurately. Besides, heavy computation is needed so that these cannot run in real-time practical systems. In this...
Neural implicit methods have achieved high-quality 3D object surfaces under slight specular highlights. However, high reflections (HSR) often appear in front of target objects when we capture them through glasses. The complex ambiguity these scenes violates the multi-view consistency, then makes it challenging for recent to reconstruct correctly. To remedy this issue, present a novel surface reconstruction framework, NeuS-HSR, based on neural rendering. In is parameterized as an signed...
Existing Neural Radiance Fields (NeRF) methods suffer from the existence of reflective objects, often resulting in blurry or distorted rendering. Instead calculating a single radiance field, we propose multi-space neural field (MS-NeRF) that represents scene using group feature fields parallel sub-spaces, which leads to better understanding network toward and refractive objects. Our scheme works as an enhancement existing NeRF methods, with only small computational overheads needed for...