A5100398583- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Image Processing Techniques and Applications
- Computer Graphics and Visualization Techniques
- Face recognition and analysis
- Video Surveillance and Tracking Methods
- Advanced Image and Video Retrieval Techniques
- Image Enhancement Techniques
- Face and Expression Recognition
- 3D Shape Modeling and Analysis
- Biometric Identification and Security
- Image and Signal Denoising Methods
- Video Coding and Compression Technologies
- Generative Adversarial Networks and Image Synthesis
- Vehicle License Plate Recognition
- Human Pose and Action Recognition
- Fatigue and fracture mechanics
- Sparse and Compressive Sensing Techniques
- Robotics and Sensor-Based Localization
- Advanced Image Fusion Techniques
- Video Analysis and Summarization
- Visual Attention and Saliency Detection
- Handwritten Text Recognition Techniques
- Turbomachinery Performance and Optimization
- Medical Image Segmentation Techniques
Xidian University
2019-2025
Hefei National Center for Physical Sciences at Nanoscale
2025
University of Science and Technology of China
2025
Hangzhou Medical College
2025
Zhejiang Provincial People's Hospital
2025
Wayne State University
2019-2024
Stanford University
2024
Xi'an Jiaotong University
2023-2024
Hainan University
2024
Dalian University of Technology
2015-2024
Deep Neural Networks (DNNs) have substantially improved the state-of-the-art in salient object detection. However, training DNNs requires costly pixel-level annotations. In this paper, we leverage observation that image-level tags provide important cues of foreground objects, and develop a weakly supervised learning method for saliency detection using only. The Foreground Inference Network (FIN) is introduced challenging task. first stage our method, FIN jointly trained with fully...
This paper reviews the 2nd NTIRE challenge on single image super-resolution (restoration of rich details in a low resolution image) with focus proposed solutions and results. The had 4 tracks. Track 1 employed standard bicubic downscaling setup, while Tracks 2, 3 realistic unknown downgrading operators simulating camera acquisition pipeline. were learnable through provided pairs high train images. tracks 145, 114, 101, 113 registered participants, resp., 31 teams competed final testing...
We present a detail-driven deep neural network for point set upsampling. A high-resolution is essential point-based rendering and surface reconstruction. Inspired by the recent success of image super-resolution techniques, we progressively train cascade patch-based upsampling networks on different levels detail end-to-end. propose series architectural design contributions that lead to substantial performance boost. The effect each technical contribution demonstrated in an ablation study....
Recent deep learning approaches to single image superresolution have achieved impressive results in terms of traditional error measures and perceptual quality. However, each case it remains challenging achieve high quality for large upsampling factors. To this end, we propose a method (ProSR) that is progressive both architecture training: the network upsamples an intermediate steps, while process organized from easy hard, as done curriculum learning. obtain more photorealistic results,...
In the last five years, edge computing has attracted tremendous attention from industry and academia due to its promise reduce latency, save bandwidth, improve availability, protect data privacy keep secure. At same time, we have witnessed proliferation of AI algorithms models which accelerate successful deployment intelligence mainly in cloud services. These two trends, combined together, created a new horizon: Edge Intelligence (EI). The development EI requires much both computer systems...
In this paper, we propose a new image super-resolution (SR) approach based on convolutional neural network (CNN), which jointly learns the feature extraction, upsampling, and high-resolution (HR) reconstruction modules, yielding completely end-to-end trainable deep CNN. However, directly training such in an fashion is challenging, takes longer time to converge may lead sub-optimal results. To address issue, train ensemble of shallow networks. The with weaker learning capability restores main...
Existing deep learning approaches to single image super-resolution have achieved impressive results but mostly assume a setting with fixed pairs of high resolution and low images. However, robustly address realistic upscaling scenarios where the relation between images is unknown, blind required. To this end, we propose solution that relies on three components: First, use degradation aware SR network synthesize HR given corresponding blur kernel. Second, train kernel discriminator analyze...
Unsupervised learning of 3D-aware generative adversarial networks (GANs) using only collections single-view 2D photographs has very recently made much progress. These 3D GANs, however, have not been demonstrated for human bodies and the generated radiance fields existing frameworks are directly editable, limiting their applicability in downstream tasks. We propose a solution to these challenges by developing GAN framework that learns generate or faces canonical pose warp them an explicit...
One impressive advantage of convolutional neural networks (CNNs) is their ability to automatically learn feature representation from raw pixels, eliminating the need for hand-designed procedures. However, recent methods single image super-resolution (SR) fail maintain this advantage. They utilize CNNs in two decoupled steps, i.e., first upsampling low resolution (LR) high (HR) size with techniques (e.g., bicubic interpolation), and then applying on upsampled LR reconstruct HR results. In...
Neural radiance fields (NeRFs) have demonstrated state-of-the-art performance for 3D computer vision tasks, including novel view synthesis and shape reconstruction. However, these methods fail scattering medium, such as haze, is prevalent in the scene. To address this challenge, we introduce DehazeNeRF a framework that robustly operates hazy conditions. extends volume rendering equation by adding physically realistic terms model atmospheric scattering. By parameterizing using suitable...
Geometric constraints are shown to enforce scale consistency and remedy the ambiguity issue in self-supervised monocular depth estimation. Meanwhile, scale-invariant losses focus on learning relative depth, leading accurate prediction. To combine best of both worlds, we learn scale-consistent a manner. Towards this goal, present scale-aware geometric (SAG) loss, which enforces through point cloud alignment. Compared prior arts, SAG loss takes into consideration during motion estimation,...
In existing deep network-based image super-resolution (SR) methods, each network is only trained for a fixed upscaling factor and can hardly generalize to unseen factors at test time, which non-scalable in real applications. To mitigate this issue, paper proposes resolution-aware (RAN) simultaneous SR of multiple factors. The key insight that essentially different but also shares common operations. attain stronger generalization across factors, we design an upsampling (U-Net) consisting...
Autonomous electric vehicle (AEV) services leverage advanced autonomous driving and technologies to provide innovative, driverless transportation solutions. The biggest challenge faced by AEVs is the limited number of charging stations long times. A critical maximizing passenger travel satisfaction while reducing AEV idle time. This involves coordinating transport tasks via leveraging information from stations, transport, data. There are four important contributions in this paper. Firstly,...