A5100687514- Image Enhancement Techniques
- Advanced Image Fusion Techniques
- Advanced Image Processing Techniques
- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- Advanced Neural Network Applications
- Fire Detection and Safety Systems
- Image and Video Quality Assessment
- Image and Signal Denoising Methods
- Computer Graphics and Visualization Techniques
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Time Series Analysis and Forecasting
- Microplastics and Plastic Pollution
- Wildlife-Road Interactions and Conservation
- Visual Attention and Saliency Detection
- Recycling and Waste Management Techniques
- Sleep and Work-Related Fatigue
- Face recognition and analysis
- Cloud Computing and Remote Desktop Technologies
- Anomaly Detection Techniques and Applications
- Cell Image Analysis Techniques
- Facial Nerve Paralysis Treatment and Research
- Writing and Handwriting Education
- Advanced Optical Imaging Technologies
National Taiwan University
2018-2025
National Yunlin University of Science and Technology
2024
Stanford University
2023-2024
National Chi Nan University
2024
National Yang Ming Chiao Tung University
2023
National Chung Hsing University
2022-2023
University of Würzburg
2023
Shanghai Artificial Intelligence Laboratory
2022
Asus (Taiwan)
2021
Asus (China)
2021
Snow is a highly complicated atmospheric phenomenon that usually contains snowflake, snow streak, and veiling effect (similar to the haze or mist). In this literature, we propose single image desnowing algorithm address diversity of particles in shape size. First, better represent complex shape, apply dual-tree wavelet transform loss network. Second, hierarchical decomposition paradigm our network for under-standing different sizes particles. Last, novel feature called contradict channel...
In this paper, an ill-posed problem of multiple adverse weather removal is investigated. Our goal to train a model with ‘unified’ architecture and only one set pretrained weights that can tackle types weathers such as haze, snow, rain simultaneously. To end, two-stage knowledge learning mechanism including collation (KC) examination (KE) based on multi-teacher student proposed. At the KC, network aims learn comprehensive bad from well-trained teacher networks where each them specialized in...
In this paper, we proposed a novel haze removal algorithm based on new feature called the patch map. Conventional patch-based algorithms (e.g. Dark Channel prior) usually performs dehazing with fixed size. However, it may produce several problems in recovered results such as oversaturation and color distortion. Therefore, designed an adaptive automatic size selection model Patch Map Selection Network (PMS-Net) to select corresponding each pixel. This network is convolutional neural (CNN),...
Images captured in a hazy environment usually suffer from bad visibility and missing information. Over many years, learning-based handcrafted prior-based dehazing algorithms have been rigorously developed. However, both exhibit some weaknesses terms of haze removal performance. Therefore, this work, we proposed the patch-map-based hybrid learning DehazeNet, which integrates these two strategies by using technique involving patch map bi-attentive generative adversarial network. In method,...
Underwater image enhancement is an important low-level computer vision task for autonomous underwater vehicles and remotely operated to explore understand the environments. Recently, deep convolutional neural networks (CNNs) have been successfully used in many problems, so does enhancement. There are deep-learning-based methods with impressive performance enhancement, but their memory model parameter costs hindrances practical application. To address this issue, we propose a lightweight...
This paper reports on the NTIRE 2022 challenge perceptual image quality assessment (IQA), held in conjunction with New Trends Image Restoration and Enhancement workshop (NTIRE) at CVPR 2022. is to address emerging of IQA by processing algorithms. The output images these algorithms have completely different characteristics from traditional distortions are included PIPAL dataset used this challenge. divided into two tracks, a full-reference track similar previous new that focuses no-reference...
This work reviews the results of NTIRE 2023 Challenge on Image Shadow Removal. The described set solutions were proposed for a novel dataset, which captures wide range object-light interactions. It consists 1200 roughly pixel aligned pairs real shadow free and affected images, captured in controlled environment. data was white-box setup, using professional equipment lights acquisition sensors. challenge had number 144 participants registered, out 19 teams compared final ranking. extend...
This report introduces two high-quality datasets Flickr360 and ODV360 for omnidirectional image video super-resolution, respectively, reports the NTIRE 2023 challenge on 360° super-resolution. Unlike ordinary 2D images/videos with a narrow field of view, can represent whole scene from all directions in one shot. There exists large gap between image/video both degradation restoration processes. The is held to facilitate development super-resolution by considering their special...
This paper introduces a novel benchmark for efficient up-scaling as part of the NTIRE 2023 Real-Time Image Super-Resolution (RTSR) Challenge, which aimed to upscale images from 720p and 1080p resolution native 4K (×2 ×3 factors) in real-time on commercial GPUs. For this, we use new test set containing diverse ranging digital art gaming photography. We assessed methods devised SR by measuring their runtime, parameters, FLOPs, while ensuring minimum PSNR fidelity over Bicubic interpolation....
This paper presents a review of the NTIRE 2023 challenge on night photography rendering.The goal was to find solutions that process raw camera images taken in nighttime conditions conditions, and thereby produce photo-quality output standard RGB (sRGB) space.Unlike previous year's competition, participants were not provided with large training dataset for target sensor.Instead, this time they given color checker illuminated by known light source.To evaluate results, sufficient number viewers...
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...
In this paper, we summarize the 1st NTIRE challenge on stereo image super-resolution (restoration of rich details in a pair low-resolution images) with focus new solutions and results. This has 1 track aiming at problem under standard bicubic degradation. total, 238 participants were successfully registered, 21 teams competed final testing phase. Among those participants, 20 submitted results PSNR (RGB) scores better than baseline. establishes benchmark for SR.
This paper reviews the NTIRE 2022 challenge on night photography rendering. The solicited solutions that processed RAW camera images captured in scenes to produce a photo-finished output image encoded standard RGB (sRGB) space. Given subjective nature of this task, proposed were evaluated based mean opinions viewers asked judge visual appearance results. Michael Freeman, world-renowned photographer, further ranked with highest opinion scores. A total 13 teams competed final phase challenge....
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...
Accidents caused by fatigue occur frequently, and numerous scholars have devoted tremendous efforts to investigate methods reduce accidents fatigued driving. Accordingly, the assessment of spirit status driver through eyes blinking frequency measurement physiological signals emerged as effective methods. In this study, a drowsiness detection system is proposed combine LF/HF ratio from heart rate variability (HRV) photoplethysmographic imaging (PPGI) percentage eyelid closure over pupil time...
Crowd counting has recently attracted significant attention in the field of computer vision due to its wide applications image understanding. Numerous methods have been proposed and achieved state-of-the-art performance for real-world tasks. However, existing approaches do not perform well under adverse weather such as haze, rain, snow since visual appearances crowds scenes are drastically different from those images clear typical datasets. In this paper, we propose a method robust crowd...
Image restoration aims to recover content from inputs degraded by various factors, such as adverse weather, blur, and noise. Perceptual Restoration (PIR) methods improve visual quality but often do not support downstream tasks effectively. On the other hand, Task-oriented (TIR) focus on enhancing image utility for high-level vision tasks, sometimes compromising quality. This paper introduces UniRestore, a unified model that bridges gap between PIR TIR using diffusion prior. The prior is...
The terrestrial environment is a significant source of anthropogenic debris emissions. While most studies on focus the marine environment, our research delves into effects human activity ingestion by studying carcasses feral pigeons. From January to June 2022, we collected gastrointestinal tracts (GI tracts) 46 pigeon in Taipei, Taiwan's capital city. results revealed that 224 samples were found, with dominant form being fibers (71.9%), which are primarily black (29.9%). Fourier transform...
In image processing, smoke may degrade visibility and deteriorate the performance of high-level vision applications. Therefore, single removal is crucial for computer vision. Currently, existing algorithms mainly leverage handcrafted priors. Moreover, these methods usually apply haze to perform due similarity between haze. However, cannot sufficiently address degradation thick suffer from residual color distortion problems non-global non-homogeneous distribution smoke. this paper, solve...
Images acquired from digital cameras are usually interfered by smoke, which may degrade the performance of object detection. There few algorithms focused on smoke removal for still images so far and we use haze to remove instead. However, there exist some differences between (e.g. particle properties localization). Thus, a dehaze algorithm has limited removal. In this paper, propose novel based machine learning detection techniques. Moreover, observed that intensity distributions not same...
Depth guided any-to-any image relighting aims to generate a relit from the original and corresponding depth maps match illumination setting of given its map. To best our knowledge, this task is new challenge that has not been addressed in previous literature. address issue, we propose deep learning-based neural Single Stream Structure network called S3Net for relighting. This an encoder-decoder model. We concatenate all images as input feed them into The decoder part contains attention...
Abstract Understanding an opposing player's behaviours and weaknesses is often the key to winning a badminton game. This study presents system extract game data from broadcast videos, visualize extracted help coaches players develop effective tactics. Specifically, we apply state‐of‐the‐art machine learning methods partition video into segments, in which each segment shows rally. Next, detect players' feet frame transform player positions court coordinate system. Finally, hit frames rally,...
Image relighting aims to recalibrate the illumination setting in an image. In this paper, we propose a deep learning-based method called multi-modal bifurcated network (MB-Net) for depth guided image relighting. That is, given and corresponding maps, new with illuminant angle color temperature is generated by our network. This model extracts features encoder. To use two effectively, adopt dynamic dilated pyramid modules decoder. Moreover, increase variety of training data, novel data process...
Single-image shadow removal aims to remove undesired information from captured images. With the development of deep convolutional neural networks, several methods have been proposed achieve promising performance in removal. However, they still struggle with limited due non-homogeneous intensity distribution shadow. To address this issue, we propose a two-stage architecture based on transformer called TSRFormer. The is divided into and content refinement networks. These two stages adopt...
Vehicle re-identification (ReID) has attracted considerable attention in computer vision. Although several methods have been proposed to achieve state-of-the-art performance on this topic, re-identifying vehicle foggy scenes remains a great challenge due the degradation of visibility. To our knowledge, problem is still not well-addressed so far. In paper, address problem, we propose novel training framework called Semi-supervised Joint Defogging Learning (SJDL) framework. First, fog removal...