陳建志

Zero-Shot Cross-Material Ptychographic Phase Reconstruction Using Deep Learning

智慧計算與科技研究所・共同指導：曾煜棋

疊層成像、4D-STEM、相位重建、深度學習、零樣本跨材料泛化、Ptychography、phase reconstruction、deep learning、zero-shot cross- material generalization

From Global to Dense Reasoning: Rearchitecting Attention for Open-Vocabulary Semantic Segmentation

智慧系統與應用研究所・共同指導：曾煜棋

視覺語言模型、開放詞彙語意分割、免訓練方法、密集視覺語言推論、Vision-Language Models、Open-Vocabulary Semantic Segmentation、Training-Free Method、Dense vision-language inference

Disentangling Human and Environmental Dynamics for Cross-Environment Wi-Fi Human Sensing

智慧系統與應用研究所・共同指導：曾煜棋

WiFi 感知、人體姿態估計、通道狀態資訊（CSI）、環境不變性學習、領域泛化、WiFi sensing、Human Pose Estimation、Channel State Information、Environment-invariant Learning、Domain Generalization

DynoDINO: Harnessing Dynamic Latent Information from DINO Features for Multi-Phase Medical Image Segmentation

智慧計算與科技研究所・共同指導：曾煜棋

多期相對比增強電腦斷層掃描（CECT）、醫學影像分割、多期相融合、影像對齊、混合注意力機制、自適應閘門機制、Multi-phase contrast-enhanced computed tomography (CECT)、Medical image segmentation、Multi-phase fusion、Image alignment、Mix-attention mechanism、Adaptive gating mechanism

Weakly Supervised Sentence-Level Alignment of Continuous Taiwanese Sign Language Videos and Text via Gaussian Prior Augmented Viterbi Learning

智慧與綠能產學研究所・共同指導：曾煜棋

臺灣手語、弱監督學習、高斯時間先驗、維特比演算法、多模態對齊、時間分割、Taiwanese Sign Language、Weakly Supervised Learning、Gaussian Temporal Prior、Viterbi Algorithm、Multimodal Alignment、Temporal Segmentation

DALSE: Drift-Aware Latent State Estimation for Federated Learning with Dynamic Participation

智能系統研究所・共同指導：曾煜棋

聯邦學習、動態客戶端參與、潛在狀態估計、客戶端漂移、非獨立同分佈資料、Federated Learning、FedDynamic Client Participation、Latent State Estimation、Client Drift、Non-IID Data

An LLM-Driven Adaptive Safety Inspection Robot for Industrial Environments via Zero- and Few-Shot Human-in-the-Loop Teaching

智慧系統與應用研究所

大型語言模型、零樣本學習、少樣本學習、工業安全巡檢、語言互動教學、LLM、Zero-shot Learning、Few-shot Learning、Industrial Safety Inspection、Language Interactive Teaching

Tiny-OAD: Online Detection of Tiny Actions in Video

智慧系統與應用研究所・共同指導：曾煜棋

線上動作偵測、微小動作辨識、小尺度動作、影片理解、時空建模、Online Action Detection、Tiny Action Recognition、Small-Scale Actions、Video Understanding、Spatiotemporal Modeling

Dual-Branch Residual Network with Mask-Guided Attention for Fungal Classification

智慧系統與應用研究所

真菌物種分類、掩模引導雙分支網路、分割驅動的空間先驗、卷積神經網路、醫學影像分析、fungal species classification、mask-guided dual-branch network、segmentationdriven spatial priors、convolutional neural networks、medical image analysis

WiFi2Caption: A Contrastive Alignment Framework for Generating Semantic Descriptions from WiFi CSI

智慧系統與應用研究所・共同指導：曾煜棋

CSI、隱私保護、人體動作語意生成、視覺知識遷移、對比式學習、Privacy Protection、Human Action Semantic Generation、Visual Knowl- edge Transfer、Contrastive Learning

Multi-Modality Fusion For Indoor Localization

智慧計算與科技研究所・共同指導：曾煜棋

多模態、室內定位、人工智慧、深度學習、Multi-Modality、Indoor Localization、Artificial Intelligence、Deep Learning

AlloEgo-VLM: Disambiguating Allocentric and Egocentric Reference Frames in Vision-Language Models

智慧計算與科技研究所・共同指導：曾煜棋

視覺語言模型、空間歧義、參考框架、自我中心/他物中心、多模態推理、Vision-Language Models、Spatial Ambiguity、Reference Frames、Egocentric/Allocentric、Multimodal Reasoning

No Label, No Problem: Semantic Labeling for Open-Set Unsupervised Video Domain Adaptation

智慧與綠能產學研究所・共同指導：曾煜棋

影片理解、動作辨識、開放集合識別、無監督領域自適應、跨領域泛化、Video Action Recognition、Open-Set、Unsupervised Domain Adaptation、Cross-Domain Generalization

Multi-Modal Pre-Training for Vehicle Trajectory Reconstruction and Localization in Simulated Urban Environments

智能系統研究所・共同指導：曾煜棋

多模態對齊、對比式學習、共享嵌入空間、軌跡檢索、CARLA 模擬城市環境、Multimodal Alignment、Contrastive Learning、Shared Embedding Space、Trajectory Retrieval、CARLA Simulated Urban Environments

Secure and Scalable Multi-Key Image Steganography with Key Isolation

智慧系統與應用研究所・共同指導：曾煜棋

隱寫、深度學習、卷積神經網路、資訊安全、金鑰隔離、Stegonagraphy、Deep Learning、CNN、Information Security、Key Isolation

Resource-Efficient V2X Federated Continual Learning from Unlabeled Streaming Data via Bandwidth-Saving Selection

智慧與綠能產學研究所・共同指導：曾煜棋

資料選擇、節省頻寬、車聯網、自監督學習、聯邦學習、持續學習、Data Selection、Bandwidth-Saving、Vehicle-to-Everything、Self-Supervised Learning、Federated Learning、Continual Learning

HCT-SD: A Hybrid 3D–2D Convolutional Transformer with Self-Distillation for Predicting Postoperative Pneumothorax Severity from Preoperative CT Images in CT-Guided Lung Biopsy

智慧計算與科技研究所・共同指導：曾煜棋

CT引導經皮肺活檢、癥狀預測、深度學習、計算機視覺、氣胸、活檢、prebiopsy CT imaging、Prediction of pneumothorax、lung tumor biopsy、deep learning、computer vision

DaMO: A Data-Efficient Multimodal Orchestrator for Temporal Reasoning with Video LLMs

智慧計算與科技研究所・共同指導：曾煜棋

影片語言模型、時間推理、多模態融合、時序感知、時間定位、計算效率優化、資料高效率訓練、Video LLM、Temporal Reasoning、Multimodal Fusion、Temporal Awareness、Temporal Localization、Computational Efficiency Optimization、Data-Efficient Training

Style-Based GAN for High Fidelity One Shot Head Reenactment

智慧系統與應用研究所・共同指導：曾煜棋

StyleGAN、人頭重建、人臉生成、電腦視覺、人工智慧、head reenactment、talking head generation、computer vision、deep learning、AI

CCKD: A semi-supervised learning method for training conditional diffusion models

智能系統研究所

知識蒸餾、擴散模型、半監督學習、隱私保護、knowledge distillation、diffusion models、semi-supervised learning、privacy preservation

HAR from Moving Cameras: A Skeleton-Oriented Multi-Modal Model

智慧計算與科技研究所・共同指導：曾煜棋

資料合成、人體動作識別、移動攝像機、姿態估計、多模態模型、data synthesis、human action recognition、moving camera、pose estimation、multi-modal model

Integrating IMU Sensors and Deep Learning for Magnetic Field Frame Transformation in Indoor Localization

智慧計算與科技研究所・共同指導：曾煜棋

室內定位、地磁序列、姿態識別、坐標系轉換、深度學習、慣性測量感測器、Indoor Localization、Geomagnetic Sequences、Pose Recognition、Frame Transformation、Deep Learning、Inertial Measurement Sensors

Mura Defect Detection: An Ensemble Approach

智慧計算與科技研究所・共同指導：曾煜棋

缺陷檢測、少樣本學習、集成學習、顯示均勻性問題、Defect detection、Few-shot learning、Ensemble learning、Mura

WiFi2Seg: Adapting WiFi CSI to Human Segmentation via Hybrid Attentions

智慧計算與科技研究所・共同指導：曾煜棋

注意力機制、人體追蹤、transformer、WiFi CSI、無線感測、attention、human tracking、wireless sensing

DirectFPN: A Parallel Approach for Efficient Feature Pyramid Construction in Temporal Action Localization

智能系統研究所・共同指導：曾煜棋

時序動作檢測、影片理解、特徵金字塔、卷積神經網路、影像辨識、電腦視覺、Temporal Action Localization、Temporal Action Detection、Video Understanding、Feature Pyramid Networks、Convolutional Neural Network、Pattern Recognition、Computer Vision

Transforming Simple Human Segmentation into Vivid Motion and Lifelike Images

智慧計算與科技研究所・共同指導：曾煜棋

簡單分割、動態控制、外觀控制、生成對抗網路、影像生成、風格控制、simple segmentation、motion control、appearance control、GAN、image generation、style control

Integrating 3D Time-of-Flight Cameras for Enhanced Image Generation and Appearance Manipulation

智慧計算與科技研究所・共同指導：曾煜棋

飛行時間、生成對抗網路、動作控制、風格控制、隱私、ToF、GAN、motion control、style control、privacy

The Survey Hole Inpainting Problem: A Machine Learning Approach

智慧與綠能產學研究所・共同指導：曾煜棋

對抗式網路、眾包、室內定位、填補、遠端感測、Adversarial Network、Crowdsourcing、Indoor Localization、Inpainting、Remote Sensing

Cross-Target Learning for Organ Image Segmentation

智慧計算與科技研究所・共同指導：曾煜棋、郭柏村

注意力機制、計算機斷層掃描、圖像分割、醫學影像、模型泛化、attention mechanism、computed tomography、mage segmentation、medical imaging、model generalization ii

Where Is My Cat? 8.6 Meters on Your Right: Learning the Geometric Displacement between WiFi

智慧計算與科技研究所・共同指導：曾煜棋

眾包、指紋、室內定位、慣性感測、多模態模型、機器人、WiFi、crowdsourcing、fingerprint、indoor localization、inertial sensing、multimodal model、robotics

Turning Trajectory Tracking into Object Detection: Vehicle Counting at Intersections by Time-Spatial Images

智慧計算與科技研究所・共同指導：曾煜棋

智慧交通系統、車輛追蹤、車輛計數、軌跡偵測、時間-空間影像、自注意力機制、Intelligent Traffic System、Vehicle Tracking、Vehicle Counting、Trajectory Detection、Time-Spatial Image、Attention

Realizing Low-Dose Medical Imaging for Radiation Therapy with Artificial Intelligence

智慧科學暨綠能學院博士班・共同指導：曾煜棋

循環長短期記憶、多體素流、低劑量電腦斷層、Circular LSTM、MVPF、Low-dose CT

Revisiting Skeleton-based Action Recognition from First-Person View

智慧計算與科技研究所・共同指導：曾煜棋

電腦視覺、資料集合成、姿態估計、姿態分析、人體動作辨識、computer vision、dataset creation、pose estimation、pose analysis、human action recognition

DIT-GAN for Synthesizing Treatment CT from Pre-treatment CT by Conditioned CBCT

光電系統研究所・共同指導：楊秉純

錐束計算機斷層掃描、預處理計劃斷層掃描、對抗式生成模型、圖像引導放 射治療、CBCT、Planning CT、GAN、IGRT

Combined Bayesian and RNN-based Hyperparameter Optimization for Efficient Model Selection Applied for autoML

智慧系統與應用研究所・共同指導：曾煜棋、郭柏村

自動模型選擇器、貝氏優化、超參數優化、自動化機器學習、遞歸類神經網路、Auto model selection (AMS)、Bayesian optimization (BO)、hyperparameter optimization (HPO)、autoML (automated machine learning)、Recurrent Neural Network (RNN)

Representative Data Selection for Continual Model Retraining

智慧與綠能產學研究所・共同指導：曾煜棋、郭柏村

生成對抗網、連續學習、資料管理、變分自編碼器、Adversarial network、Continual learning、Incremental learning、Data management、Variational autoencoder

Natural Language-based Road Video Retrieval

智慧與綠能產學研究所・共同指導：曾煜棋

智慧交通系統、車輛檢索、自然語言處理、時間-空間影像、對比式學習、Intelligent Traffic System、Vehicle retrieval、Natural Language Processing、Time-Spatial Image、Contrastive Learning

Realistic Pedestrian Shadow Generation by 2D-to-3D Object-Lifting

智慧與綠能產學研究所

戶外場景生成、行人生成、物件立體化、影子生成、Outdoor Scene Synthesis、Human Synthesis、Object Lifting、Shadow Generation

Zero In-map Learning of WiFi Fingerprints

智慧計算與科技研究所・共同指導：曾煜棋

對抗式神經網路、室內定位、填補、WiFi指紋庫、無線訊號、Adversarial network、Indoor localization、Inpainting、WiFi fingerprint、Wireless sensing

Zero In­map Learning of WiFi Fingerprints

智慧計算與科技研究所・共同指導：曾煜棋

生成對抗網路、眾包、室內定位、插補、無線網路指紋庫、無線傳感、Adversarial network、Crowdsourcing、Indoor localization、Inpainting、WiFi fingerprint、Wireless sensing

Low-Resolution Thermal Sensor-Guided Image Synthesis

智慧與綠能產學研究所・共同指導：張明峰

感測器、熱影像、生成模型、影像生成、生成對抗網路、Sensors、Thermal Imaging、Generative Model、Image Synthesis、Generative Adversarial Network

Graph Attention Transformer for Unsupervised Multivariate Time Series Anomaly Detection

智慧與綠能產學研究所・共同指導：曾煜棋

異常檢測、圖神經網路、時間序列預測、多特徵時間序列、變壓器、Anomaly detection、Graph Neural Network、GRU­forecasting、Multivariate time series data、Transformer

A Light-Weight Framework for Multi-movement Vehicle Counting

智慧計算與科技研究所・共同指導：曾煜棋

智慧交通系統、多路徑車輛計數、時間-空間影像、物件偵測、度量學習、Intelligent Traffic System、Multi-movement Vehicle Counting、Time-Spatial Image、Object Detection、Metric Learning

MR-Based UAV Route Planning for the Coverage Task

智慧計算與科技研究所・共同指導：曾煜棋

混合實境、HoloLens、無人機、路徑規畫、Mixed Reality、Unmanned Aerial Vehicle、Route Planning

A Deep Learning Approach to the Boundary Localization Problem

智慧系統與應用研究所

指紋定位、室內外邊界區域、深度學習、自動編碼器和長短期記憶模型、多傳感器和多任務、現實生活中的實驗、Fingerprint positioning、indoor and outdoor boundary areas、deep learning、auto-encoder and LSTM、multi-sensor and multi-task、real-life experiments

PDL3D: 3D Attention Module with Partial Dense Layer for Small-to-Medium Dataset on Object Detection

智慧與綠能產學研究所・共同指導：曾煜棋、郭柏村

深度學習、注意力機制、瑕疵檢測、Deep learning、Attention、Defect detection

Knowledge Transfer Learning from Classification to Segmentation for Scarce Medical Images

智慧計算與科技研究所・共同指導：曾煜棋、郭柏村

類激活熱力圖、深度學習、醫療圖像、語義分割、轉移學習、Class Activation Map、Deep Learning、Medical Image、Semantic Segmentation、Transfer Learning

Low-Cost Computer Vision Intelligent Transportation Solutions in 5G

智慧計算與科技研究所

自動駕駛、物聯網、深度學習、時間-空間影像、車輛計數、第五代行動通訊技術、Autonomous Driving、Internet-of-Things (IoT)、Deep Learning、Time-Spatial Image (TSI)、Vehicle Counting、5G