2026
EffiMiniVLM: A Compact Dual-Encoder Regression Framework
Yin-Loon Khor*, Yi-Jie Wong*, Yan Chai Hum† (* equal contribution)
IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2026 Top 4, LoViF @ CVPR 2026
A compact dual-encoder vision–language regression framework combining EfficientNet-B0 and MiniLMv2 for product quality scoring, achieving CES 0.40 with only 27.7M parameters and 6.8 GFLOPs — the lowest resource cost on the CVPR 2026 LoViF leaderboard, 4× to 8× more efficient than comparable top-5 methods, and the only top submission trained without external datasets.
EffiMiniVLM: A Compact Dual-Encoder Regression Framework
Yin-Loon Khor*, Yi-Jie Wong*, Yan Chai Hum† (* equal contribution)
IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2026 Top 4, LoViF @ CVPR 2026
A compact dual-encoder vision–language regression framework combining EfficientNet-B0 and MiniLMv2 for product quality scoring, achieving CES 0.40 with only 27.7M parameters and 6.8 GFLOPs — the lowest resource cost on the CVPR 2026 LoViF leaderboard, 4× to 8× more efficient than comparable top-5 methods, and the only top submission trained without external datasets.
2024
Cross-City Building Instance Segmentation: From More Data to Diffusion-Augmentation
Yi-Jie Wong, Yin-Loon Khor, Mau-Luen Tham, Ban-Hoe Kwan, Anissa Mokraoui, Yoong Choon Chang
IEEE International Conference on Big Data (BigData) 2024 1st Place, BEGC 2024
A data-centric approach to cross-city building instance segmentation that improves generalization through open-source data from the Microsoft Building Footprint dataset and a novel segmentation-guided diffusion augmentation pipeline. The YOLOv8-based solution achieved a private F1-score of 0.897, ranking 1st globally in the IEEE BigData Cup 2024 Building Extraction Generalization Challenge.
Cross-City Building Instance Segmentation: From More Data to Diffusion-Augmentation
Yi-Jie Wong, Yin-Loon Khor, Mau-Luen Tham, Ban-Hoe Kwan, Anissa Mokraoui, Yoong Choon Chang
IEEE International Conference on Big Data (BigData) 2024 1st Place, BEGC 2024
A data-centric approach to cross-city building instance segmentation that improves generalization through open-source data from the Microsoft Building Footprint dataset and a novel segmentation-guided diffusion augmentation pipeline. The YOLOv8-based solution achieved a private F1-score of 0.897, ranking 1st globally in the IEEE BigData Cup 2024 Building Extraction Generalization Challenge.
Multi-Task YOLO for Vehicle Colour Recognition and Automatic License Plate Recognition
Yin-Loon Khor, Yi-Jie Wong, Mau-Luen Tham, Yoong Choon Chang, Ban-Hoe Kwan, Kok-Chin Khor
IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS) 2024
A one-stage multi-task YOLOv8 framework that unifies license plate detection, OCR, and vehicle colour recognition into a single model with three task-specific heads sharing a common backbone. Achieves mAP scores of 0.778 (OCR), 0.963 (LP), and 0.881 (VCR) while running 3.07× faster than the conventional sequential single-task pipeline.
Multi-Task YOLO for Vehicle Colour Recognition and Automatic License Plate Recognition
Yin-Loon Khor, Yi-Jie Wong, Mau-Luen Tham, Yoong Choon Chang, Ban-Hoe Kwan, Kok-Chin Khor
IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS) 2024
A one-stage multi-task YOLOv8 framework that unifies license plate detection, OCR, and vehicle colour recognition into a single model with three task-specific heads sharing a common backbone. Achieves mAP scores of 0.778 (OCR), 0.963 (LP), and 0.881 (VCR) while running 3.07× faster than the conventional sequential single-task pipeline.
Automating Coastal Vulnerability Assessment: AI-Driven Geospatial Analysis via Building Damage Detection
Yin-Loon Khor, Yi-Jie Wong, Ziwei Liu
2024 Top 10 Semi-Finalist, EY 2024
An AI-driven geospatial analysis pipeline that automates coastal disaster assessment by detecting damaged buildings from satellite imagery. The model is pretrained on Microsoft's Building Footprint dataset and fine-tuned with a mix of manually and self-annotated data, then used to generate damage-count and damage-ratio heatmaps for disaster response and resilience planning. The solution ranked Top 10 Global Semi-Finalist in the 2024 EY Open Science Data Challenge out of 11,000+ registrants.
Automating Coastal Vulnerability Assessment: AI-Driven Geospatial Analysis via Building Damage Detection
Yin-Loon Khor, Yi-Jie Wong, Ziwei Liu
2024 Top 10 Semi-Finalist, EY 2024
An AI-driven geospatial analysis pipeline that automates coastal disaster assessment by detecting damaged buildings from satellite imagery. The model is pretrained on Microsoft's Building Footprint dataset and fine-tuned with a mix of manually and self-annotated data, then used to generate damage-count and damage-ratio heatmaps for disaster response and resilience planning. The solution ranked Top 10 Global Semi-Finalist in the 2024 EY Open Science Data Challenge out of 11,000+ registrants.