Featured Projects

🚀 STAC: Saliency-Guided Transformer Attention with Pixel-Level Contrastive Learning

Published in IEEE Transactions on Industrial Informatics (TII), 2026

A novel weakly supervised framework for industrial defect localization. Combines saliency-guided transformer attention and pixel-level contrastive learning to achieve high-precision defect segmentation using only image-level labels.

  • Significantly improves boundary precision and localization accuracy
  • Outperforms state-of-the-art methods on NEU-Seg, DAGM, MTD, and MVTec datasets
  • Strong generalization capability on PASCAL VOC

Links:

🚀 Semi-supervised segmentation framework designed for real-world industrial surface defect detection

SimEps: Published in IEEE Transactions on Industrial Informatics (TII), 2022 UAPS: Published in Computers in Industry (CI), 2023 Links:

🛠️ CNN-Transformer Hybrid Segmentation Model for Industrial Defect Segmentation

  • Uses transformer encoder + CNN decoder architecture
  • Multi-scale feature fusion and boundary-aware refinement
  • Achieved high performance on steel surface defect datasets (NEU-Seg & SSDD)
  • dseg-models - GitHub Repository

More Projects Coming Soon

I am currently working on several new projects in the areas of:

  • Vision-Language Models for industrial inspection
  • Anomaly detection in complex manufacturing environments
  • Lightweight models for edge deployment
  • Semantic reasoning and Chain-of-Thought (CoT)
  • Multimodal approaches and generalization