Evan's Profile

About Me

Hello! I'm Evan Ho, a master's student at National Tsing Hua University (NTHU) in Taiwan. I major in Computer Science and am part of the Computer Vision Lab. My research focuses on image anomaly detection. I'm currently participating in an exchange program at the University of Saskatchewan, Canada. I’m excited to continue learning and exploring new opportunities in the field.

This is a self introduction updates in Dec. 2024 using BootStrap, OpenLayers, Font Awesome and Chart.js.

Project

InstAD: Instance-aware Segmentation Framework for Zero-shot Multi-instance Anomaly Detection

Multi-instance anomaly detection is a crucial task but received little attention. In real-world applications, detection scenarios are often not conducted under perfectly aligned conditions, with multiple instances potentially appearing in a single shot. To address this problem, we refine the failure segmentation maps provided by the SAM model and propose a few-shot/zero-shot framework for multi-instance anomaly detection and segmentation. (July, 2024)
(Accepted to IEEE ICASSP 2025)

Diffusion-based Synthetic Anomalies for Industrial Inspection

This work tries to address the scarcity issue of defect images in industrial anomaly detection field by leveraging diverse defects generated through diffusion models. The synthetic dataset is used for performance evaluation in anomaly detection and segmentation tasks. These images are acquired by crawling and organizing similar ones from the web. (July, 2023)

Historical Typhoon Search Engine Based on Track Similarity

We developed a historical typhoon search engine based on track similarity. Building on previous work, we introduced the "Recentness Dominance Principle" to enhance the similarity weighting. The information display panel offers a concise and user-friendly interface, enabling decision-makers to quickly review historical typhoons. This provides a rapid method for understanding the current situation by leveraging insights from past events. (Oct., 2019)
(Accepted to MDPI 2019)

Works

Conference Paper

C.Y. Ho, S.H. Lai (2025, April). InstAD: Instance-aware Segmentation Framework for Zero-shot Multi-instance Anomaly Detection. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE.
Lin, S. C., Lee, H. W., Hsieh, Y. H., Ho, C. Y., & Lai, S. H. (2023). Masked Attention ConvNeXt Unet with Multi-Synthesis Dynamic Weighting for Anomaly Detection and Localization[LINK]
Hsieh, C.-M., Ho, C.-Y., Kung, H.-K., Chan, H.-Y., Tsai, M.-H. and Tsai, Y.-C. (2020). Track Similarity-based Typhoon Search Engine for Disaster Preparedness. 2020 Proceedings of the 37th ISARC, Kitakyushu, Japan. October 27-29. doi:10.22260/ISARC2020/0188[LINK]

Workshop

2nd Place in MultiSports Challenge, ECCV 2022 DeeperAction Challenge and Workshop on Detailed Video Action Understanding and Anomaly Recognition[LINK]

Journal Paper

Kung, H.-K. , Hsieh, C.-M., Ho, C.-Y., Tsai, Y.-C. Tsai, M.-H., Chan, H.-Y. (2021) Data-Augmented Hybrid Named Entity Recognition for Disaster Management by Transfer Learning, Computing and Artificial Intelligence, Appl. Sci. 2020, xx, 5; doi:10.3390/appxx010005[LINK]
Tsai, M.-H., Chan, H.-Y., Hsieh, C.-M., Ho, C.-Y., Kung, H.-K., Tsai, Y.-C. and Cho, I.-C. (2019). Historical Typhoon Search Engine Based on Track Similarity. International Journal of Environmental Research and Public Health, 16(24), 4879. doi:10.3390/ijerph16244879[LINK]