2022/12 - Present
專任研究助理
國立成功大學統計學系
- 專案管理: 管理 5-8 人團隊，協助醫院醫療數據分析、清理與建模前處理。
- 專案執行: 執行國科會計畫，產出論文、短文與研究報告。
- 競賽參與: 參與多個領域內與跨領域競賽，並取得良好成績。
- 協助指導: 協助同仁解決實驗室研究與技術難題。
#深度學習模型 #撰寫論文 #整合管理 #研究報告與計畫執行
個人介紹
關於我
以研究落地為核心，把 DeepFake、醫療、遙測與文字資料轉成可驗證、可部署的 AI 系統。
I focus on turning deepfake, medical, remote sensing, and text data into verifiable AI systems that can move from research to implementation.
ACVLAB 團隊於 IJCAI-ECAI 2026 DDL 2.0 Workshop 的 DDL-X Challenge Track 3 取得最終第四名；提出 TRACE 深度偽造定位方法，論文獲 workshop 接受。
我於 2018 年取得屏東科技大學資訊管理學系學士學位，並於 2020 年完成碩士學位。我的研究興趣集中於計算機視覺、深度學習、醫療影像處理和自然語言處理等領域。
2020 年至 2022 年間，我在長庚醫療財團法人進行中風 MRI 影像識別研究，研究成果已發表於 NeuroImage: Clinical。
自 2022 年起，我從事高光譜影像傳輸、醫療 NLP 與 DeepFake Detection 平台落地，成果發表於 ACM、IEEE TGRS 與 IJCV。
The ACVLAB team placed fourth in Track 3 of the DDL-X Challenge at the IJCAI-ECAI 2026 DDL 2.0 Workshop. The team proposed TRACE for auditable deepfake region localization, with the paper accepted by the workshop.
I received a bachelor's degree in Information Management from Pingtung University of Science and Technology in 2018 and completed my master's degree in 2020.
From 2020 to 2022, I conducted MRI stroke image recognition research at Chang Gung Hospital, developing segmentation and classification models published in NeuroImage: Clinical.
Since 2022, I have worked on hyperspectral image transmission, medical NLP, and deployed DeepFake Detection systems, with publications in ACM, IEEE TGRS, and IJCV.
工作經驗
工作經驗
研究、專案管理與跨領域團隊協作並行。
Research execution, project management, and cross-disciplinary collaboration.
2020/09 - 2022/11
醫研助理
長庚醫療財團法人林口長庚紀念醫院
- 計畫執行: 負責醫療影像辨識相關計畫，並將成果撰寫成論文發表。
- 醫療數據分析: 協助同仁分析醫療數據，定期分享並指導專案執行。
#Python #Linux #PyTorch #TensorFlow #Firebase
2022/12 - Present
Research Assistant
Department of Statistics, National Cheng Kung University
- Project Management: Managed a team of 5-8 members and supported hospital data analysis, cleaning, and modeling preparation.
- Project Execution: Executed National Science and Technology Council projects, producing papers, short articles, and research reports.
- Competition Participation: Joined domain-specific and interdisciplinary competitions with notable outcomes.
- Guidance and Support: Helped colleagues resolve research and technical issues in the laboratory.
#DeepLearning #PaperWriting #ProjectManagement #ResearchReports
2020/09 - 2022/11
Medical Research Assistant
Chang Gung Memorial Hospital, Linkou
- Project Execution: Led medical image recognition projects and prepared research results for publication.
- Medical Data Analysis: Supported medical data analysis and provided recurring project guidance.
#Python #Linux #PyTorch #TensorFlow #Firebase
研究主題
研究專案
現階段以 DeepFake Detection 平台落地為主，整合偵測、定位、解釋與可部署工程流程。
Current work centers on deployed DeepFake Detection platforms that combine detection, localization, explainability, and production-oriented engineering.
DeepFake Detection 平台
已落地深度偽造偵測平台，整合影像真偽分類、可疑區域定位與可解釋報告輸出。
- 支援研究模型到平台服務的部署流程。
- 延伸 TRACE 方法於可稽核的偽造區域定位。
醫療影像
專注於急性缺血性中風 MRI 影像的分割與分類。
- 開發 SGD-Net，用於準確分割加權影像中的 AIS 病灶。
- 進一步開發 SGD-Net Plus，將病灶大小分佈納入腦損傷量化分析。
高光譜影像
解決高光譜影像的高效壓縮與傳輸問題。
- 開發 RTCS，在有限資源下高效重建 HSI 影像。
- 支援遠端設備重建，減少對地面計算資源的需求。
自然語言處理
專注於醫療文本數據分類與雙語語意表示。
- 開發雙語記錄中低血壓分類模型 MMIDHC。
- 使用多語言 Sentence-BERT 進行分類任務。
大型語言模型應用
將生成式 AI 與 LLM 導入影像復原、AOI 瑕疵檢測與報告生成。
- 提升高光譜影像資料復原能力。
- 結合 AOI 技術進行工業產品瑕疵檢測與自動化標註。
DeepFake Detection Platform
Deployed a deepfake detection platform that integrates image-level classification, suspicious-region localization, and explainable report output.
- Supports the path from research models to platform services.
- Extends TRACE-style auditable localization for forged regions.
Medical Imaging
Focused on segmentation and classification of Acute Ischemic Stroke lesions in MRI images.
- Developed SGD-Net for accurate AIS lesion segmentation in weighted images.
- Extended the work with SGD-Net Plus by incorporating lesion size distribution for quantitative analysis.
Hyperspectral Imaging
Focused on efficient compression and transmission of hyperspectral images.
- Developed RTCS for efficient HSI reconstruction under limited resources.
- Enabled reconstruction for remote devices, reducing reliance on ground-based computing resources.
Natural Language Processing
Focused on medical text classification and bilingual semantic representation.
- Developed MMIDHC for bilingual intra-dialytic hypotension classification.
- Applied multilingual Sentence-BERT to classification tasks.
Large Language Model Applications
Applied generative AI and LLMs to image recovery, AOI defect detection, and report generation.
- Improved data restoration capabilities for hyperspectral imaging.
- Combined AOI workflows with automated defect annotation for industrial inspection.
能力工具
擅長技能與工具
從模型實驗、資料處理到研究成果撰寫，保留可重現的工程流程。
From model experiments and data processing to research writing, with reproducible engineering workflows.
Programming
- Python
- Git / GitHub
- Linux Ubuntu
- Docker
- AI Agent Workflows
Frameworks
- PyTorch
- TensorFlow
- Scikit-learn
- Jupyter Notebook
- DeepFake Detection
- Agentic Projects 擅長使用各項 AI Agent 進行需求拆解、程式撰寫、測試驗證與專案整合。 Experienced in using AI agents for planning, coding, testing, and project integration.
- Computer Vision Medical imaging, visual recognition, image restoration, and AOI inspection workflows.
- NLP / LLM Medical text classification, multilingual representation, and report generation.
- Research Delivery Paper writing, project execution, team guidance, and reproducible experiment tracking.
發表成果
出版論文
涵蓋 DeepFake Detection、遙測、醫療影像、醫療 NLP、車輛再辨識與 COVID-19 影像辨識。
Covering deepfake detection, remote sensing, medical imaging, medical NLP, vehicle re-identification, and COVID-19 image recognition.
-
UMCL: Unimodal-generated Multimodal Contrastive Learning for Cross-compression-rate Deepfake Detection
International Journal of Computer Vision, 134, Article 40 (2026) · DOI · arXiv
-
TRACE: Token Reference Attention for Auditable Deepfake Region Localization
IJCAI-ECAI 2026 DDL 2.0 Workshop, DDL-X Challenge Track 3 solution paper, accepted
-
Revisiting Vision-Language Features Adaptation and Inconsistency for Social Media Popularity Prediction
IEEE Transactions on Geoscience and Remote Sensing, 2024
-
Real-Time Compressed Sensing for Joint Hyperspectral Image Transmission and Restoration for CubeSat
IEEE Transactions on Geoscience and Remote Sensing, 2024
-
Bag of Tricks of Hybrid Network for COVID-19 Detection of CT Scans
IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops, 2023
-
Strong Baseline and Bag of Tricks for COVID-19 Detection of CT Scans
arXiv preprint, 2023
-
Multimodality-aware Intra-dialytic Hypotension Classifier: A Bilingual NLP Approach to Classify Dialysis Records
ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics, 2023
-
Strong Baseline for Vehicle Re-identification in the Wild
IEEE Visual Communications and Image Processing, 2019
-
Semantic Segmentation Guided Detector for Segmentation, Classification, and Lesion Mapping of Acute Ischemic Stroke in MRI Images
NeuroImage: Clinical, 2022
活動紀錄
活動照片
國際競賽、研究展示與團隊活動中的現場紀錄。
Moments from research demos, competitions, and team collaboration.