Students from the College of Computer Science at National Yang Ming Chiao Tung University (NYCU) continue to bring innovative and outstanding research to the world's leading academic stage. This issue features eight papers from different labs, selected for presentation at top international conferences including NeurIPS, CVPR, ICRA, ACL, and ICCV. Spanning fields such as image processing, computer vision, robotic automation, natural language processing, and medical artificial intelligence, these achievements not only reveal to the world the depth and breadth of NYCU's research, but also demonstrate the solid research capabilities and international competitiveness of our students. Through exchanges with researchers from around the globe, these students are sure to gain valuable insights, returning with a renewed perspective as they advance toward even greater academic achievements.

Title: BlurDM: A Blur Diffusion Model for Image Deblurring

Authors: Jin-Ting He, Fu-Jen Tsai, Yan-Tsung Peng, Min-Hung Chen, Chia-Wen Lin, Yen-Yu Lin

Advisor: Professor Yen-Yu Lin

Conference: The Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS)

Significance:

NeurIPS is one of the most influential top-tier international conferences in the fields of artificial intelligence and machine learning. In terms of academic metrics, NeurIPS has an h5-index of approximately 337 on Google Scholar Metrics and ranks first in the Artificial Intelligence category, reflecting the exceptionally high citation impact and academic influence of its papers. In addition, the conference attracts over 20,000 submissions each year, with an acceptance rate of approximately 20–25%, making competition extremely fierce. As a result, publishing research at NeurIPS is widely regarded as a mark of high innovation and international academic value.

The experience of Jin-Ting He:

I feel deeply honored to have taken part in this research and to have traveled abroad for the conference. First, I would like to give special thanks to my advisor, Professor Yen-Yu Lin, and all of my collaborators, for their support and effort throughout the entire research and submission process. What left the deepest impression on me was the rebuttal stage, during which nearly every co-author was on call around the clock. Whenever the reviewers raised new questions, we would immediately discuss them online, quickly analyze the review comments, and work together to craft the most appropriate responses. This intensive, closely coordinated process was exhausting, but seeing the paper accepted made it all worthwhile. This was also my first time attending an international academic conference in the United States, where the meeting was held. Being able to exchange ideas in person with researchers from around the world and learn about the latest research directions and results was an incredibly rare and valuable experience for me. Finally, I would like to once again thank my advisor and all my collaborators for their support and dedication in completing this research together.

 

Title: From Alignment to Reason: Multi-Agent Debate for Tactical Badminton Video Retrieval

Authors: Yi-Xiang Zhang, Yu-Shuen Wang

Advisor: Professor Yu-Shuen Wang

Conference: IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2026

Significance:

CVPR (Computer Vision and Pattern Recognition) is one of the most influential international conferences in the field of computer vision. Each year, the conference brings together researchers from academia and industry worldwide to present cutting-edge work in computer vision, image and video understanding, generative AI, 3D vision, autonomous driving, and more. Having a paper accepted at CVPR reflects strong recognition of the research's innovation, technical depth, and international impact.

The experience of Yi-Xiang Zhang:

I am very honored that our research was presented at CVPR. I would like to sincerely thank Professor Yu-Shuen Wang for his careful guidance, as well as Professor Chih-Wei Chang and Coach Chih-Chuan Wang for their valuable advice throughout the research process. This work addresses the challenges of video understanding and tactical retrieval in badminton matches by proposing a “perception-reasoning decoupled” framework. Using a multi-agent debate reasoning framework, it analyzes offensive and defensive intent as well as scoring context, allowing users to query specific tactical segments. Attending CVPR also exposed me to a wide range of cutting-edge research from different fields, and through exchanges with international scholars, I learned different ways of thinking about and approaching research.

 

Published Paper: HetroD: A High-Fidelity Drone Dataset and Benchmark for Autonomous Driving in Heterogeneous Traffic

Authors: Yu-Hsiang Chen, Wei-Jer Chang, Christian Kotulla, Thomas Keutgens, Steffen Runde, Tobias Moers, Christoph Klas, Wei Zhan, Masayoshi Tomizuka, Yi-Ting Chen

Advisor: Professor Yi-Ting Chen

Conference: IEEE International Conference on Robotics and Automation 2026 (ICRA 2026)

Significance:

The IEEE International Conference on Robotics and Automation (ICRA), hosted by the IEEE Robotics and Automation Society, is one of the most representative and influential top-tier international conferences in the fields of robotics, autonomous driving, and artificial intelligence worldwide. The conference brings together leading universities, research institutions, and technology companies to present cutting-edge research in robotic perception, control, embodied intelligence, autonomous driving, and intelligent transportation. Beyond its strong academic influence, ICRA also serves as an important platform for international academia-industry exchange, technology demonstration, and research collaboration, reflecting the latest trends and future directions in the field. Presenting research at ICRA not only signifies international peer recognition of research quality, but also helps raise the international visibility of research outcomes, build cross-border collaboration networks, and connect academic achievements with industry needs.

The experience of Yu-Hsiang Chen:

This time, I traveled to Vienna, Austria to attend ICRA 2026, where I presented “HetroD: A High-Fidelity Drone Dataset and Benchmark for Autonomous Driving in Heterogeneous Traffic.” HetroD focuses on real-world heterogeneous traffic involving cars, motorcycles, bicycles, and pedestrians, using drone-collected data and a benchmark to support trajectory prediction, traffic simulation, scenario generation, and autonomous driving validation. During the conference, many scholars and industry representatives showed strong interest in our data collection methods, application value, and future extensions; more than fifty dataset requests have already been received, confirming that this research genuinely addresses the needs of the international community. Through exchanges with teams from Waabi, UC Berkeley, UCLA, Waymo, and Toyota Research Institute, I gained further insight into the latest trends in autonomous driving and generative simulation, and came to realize that publishing a paper is only the starting point — real impact comes from whether the results continue to be used, evolve into challenges and evaluation standards, and foster cross-border collaboration. This conference also prompted me to rethink traffic simulation methods that rely solely on imitation learning; going forward, I plan to further explore road users' intentions, anticipatory behavior, and interactive dynamics. The biggest change from this experience is that it shifted my focus from completing a single study to thinking about how to leverage our lab's strengths in heterogeneous traffic data to build a research direction with long-term value and international recognition.

 

Published Paper: Safety-Aware Dialogue System for Postoperative Oral Cancer Care with Structured Clarification and a Clinically Curated Dataset

Authors: Tzu-Chi Liu, Hui-Ying Yang, Shiow-Ching Shun, Yu-Chi Chen, Lu-Yen Anny Chen, Yong-Sheng Chen

Advisor: Professor Yong-Sheng Chen

Conference: Annual Meeting of the Association for Computational Linguistics (ACL)

Significance:

The Annual Meeting of the Association for Computational Linguistics (ACL) is one of the most representative top-tier international conferences in the fields of computational linguistics and natural language processing, consistently featuring cutting-edge research on language models, natural language understanding, dialogue systems, information extraction, machine translation, and interdisciplinary applications. ACL papers undergo rigorous peer review, and acceptance at this conference reflects strong recognition of a study's methodological innovation, technical rigor, and international academic impact.

The experience of Tzu-Chi Liu:

I am honored to have presented the research from my PhD studies at ACL. This project gave me a deep appreciation that medical AI cannot simply pursue model performance or fluent responses — it must also understand the uncertainty and safety risks inherent in clinical settings. This is especially true in postoperative oral cancer care, where patients' questions are often brief and lack sufficient context, requiring the system to proactively clarify questions and respond based on trustworthy clinical knowledge. The research process also taught me the importance of interdisciplinary communication; only by combining information technology with clinical expertise can AI truly meet real-world care needs. I would like to thank my advisor, Professor Yong-Sheng Chen, and our collaborators for their support.

 

Published Paper: StealthAttack: Robust 3D Gaussian Splatting Poisoning via Density-Guided Illusions

Authors: Bo-Hsu Ke, You-Zhe Xie, Yu-Lun Liu, Wei-Chen Chiu

Advisor: Professor Wei-Chen Chiu and Professor Yu-Lun Liu

Conference: International Conference on Computer Vision (ICCV 2025)

Significance:

ICCV (International Conference on Computer Vision), jointly organized by the IEEE Computer Society and the Computer Vision Foundation (CVF), is one of the most authoritative international conferences in the field of computer vision, ranked alongside CVPR and ECCV as one of the field's three premier conferences. It is designated a CCF Recommended Category A conference, ranked A1 in Qualis and A in ERA, and attracts more than 5,000 researchers annually. ICCV 2025 was held in Honolulu, Hawaii, over five days, featuring technical paper presentations, workshops, and exhibitions in various formats. This year's acceptance rate was just 24.19%, the lowest in nearly a decade, reflecting intense competition.

The experience of Bo-Hsu Ke:

I am extremely grateful to Professor Wei-Chen Chiu and Professor Yu-Lun Liu for their guidance, which gave me the opportunity to attend this major international event. In addition to presenting our paper, I was also honored to receive the Best Poster Award at the ICCV Responsible Imaging Workshop. During the conference, I actively attended numerous workshops as well as oral presentations and poster sessions at the main conference, where I not only observed many creative and in-depth pieces of cutting-edge research, but also had the chance to see talks by leading scholars in the field such as Noah Snavely and William T. Freeman. I found these especially inspiring, particularly the perspective of “starting from a dataset to identify a research question.” At my own poster session, I explained my work to more than a dozen scholars from around the world. I was initially nervous about presenting in English, but with encouragement from my mentors I gradually overcame that anxiety, receiving a great deal of positive feedback and suggestions while getting to know researchers from different countries through meaningful academic exchange. This experience at ICCV 2025 not only benefited my academic research greatly and significantly improved my English communication skills, but more importantly, it broadened my perspective on the field of computer vision. These gains will serve as important nourishment and motivation on my future academic journey.

 

Published Paper: FIPER: Factorized Features for Robust Image Super-Resolution and Compression

Authors: Yang-Che Sun, Cheng Yu Yeo, Ernie Chu, Jun-Cheng Chen, Yu-Lun Liu

Advisor: Professor Yu-Lun Liu

Conference: NeurIPS (Neural Information Processing Systems) 2025

Significance:

NeurIPS is the world's most prestigious and largest annual academic conference for artificial intelligence and machine learning.

The experience of Yang-Che Sun:

This is my first paper and also my first top-conference experience. From brainstorming, experiments, and evaluation to writing and submission, it has been a long journey, and I have learned a lot throughout the process. This paper was rejected four times before finally being accepted. Although receiving negative reviews was frustrating each time, I persisted and kept refining the manuscript. I believe that not giving up hope is one of the most important parts of innovation and research.

 

Published Paper: GRITS: A Spillage-Aware Guided Diffusion Policy for Robot Food Scooping Tasks

Authors: Yen-Ling Tai, Yi-Ru Yang, Kuan-Ting Yu, Yu-Wei Chao, Yi-Ting Chen

Advisor: Professor Yi-Ting Chen

Conference: IEEE International Conference on Robotics and Automation (ICRA 2026)

Significance:

ICRA is one of the most representative top-tier international conferences in the field of robotics and automation. The conference covers key research topics such as robotic perception, planning, control, manipulation, human-robot interaction, and automation, bringing together representative achievements from academia, research institutions, and industry worldwide each year, with high international visibility and significant academic impact.

The experience of Yen-Ling Tai:

I was honored to give an oral presentation at ICRA 2026, sharing our spillage-aware guided diffusion policy for robot food scooping tasks. I would like to thank my advisor and collaborators for their guidance and support. During the conference, we had the chance to closely observe live demonstrations of robotic arms, humanoid robots, and quadruped robots, and through exchanges with scholars from around the world, gained a deeper understanding of the challenges of translating research into real-world applications. This experience not only broadened my international perspective, but also helped me better understand where my research stands within the fields of robot learning and manipulation, and I look forward to continuing to improve my academic communication and exchange skills in the future.

 

Published Paper: LightsOut: Diffusion-based Outpainting for Enhanced Lens Flare Removal

Authors: Shr-Ruei Tsai, Wei-Cheng Chang, Jie-Ying Lee, Chih-Hai Su, Yu-Lun Liu

Advisor: Professor Yu-Lun Liu

Conference: International Conference on Computer Vision (ICCV) 2025

Significance:

ICCV is one of the most influential academic conferences in the field of computer vision, showcasing the latest theories, techniques, and applications in areas such as image classification, object detection, 3D vision, and generative models. In 2025, a total of 11,239 papers were reviewed, of which 2,698 were accepted, resulting in an acceptance rate of approximately 24%.

The experience of Shr-Ruei Tsai:

I would like to thank Professor Yu-Lun Liu for his guidance and my teammates for their collaboration. Looking back to the start of this project, after discussing with our advisor we decided to work together toward the most challenging goal of publishing a paper, and we were fortunate enough to eventually be accepted at ICCV, something we had not dared to imagine at the outset. In this research, we found that lens flare removal relies heavily on light-source information, causing performance to drop significantly when no light source is visible in the image. To address this underlying issue, we designed a diffusion-based outpainting approach to imagine the light source beyond the image boundary, and subsequent experiments showed that our method effectively improves performance across several different flare removal models. Attending this conference was an unforgettable experience — not only did I get to see the most cutting-edge computer vision research from around the world, but it also strengthened my determination to continue exploring and deepening my expertise in the field of computer vision.