Projects

Ultra-high Resolution Image Understanding

Understanding ultra-high resolution images (i.e., >2K) is a challenging problem due to high computational burden, which can benefit the fields of remote sensing or medical imaging. The goal is to strike the balance between effectiveness and efficiency. Ultimately, we hope we can accomplish the real-time processing for ultra-high resolution images.

• From Contexts to Locality: Ultra-high Resolution Image Segmentation via Locality-aware Contextual Correlation ICCV, 2021 [pdf][arxiv][code]: we propose a novel patch-based ultra-high resolution image segmentation paradigm that correlates local regions with their contexts.
• Ultra-high Resolution Image Segmentation via Locality-aware Context Fusion and Alternating Local Enhancement
  International Journal of Computer Vision (IJCV), 2024 [arxiv]: we significantly extend and reinnovate our ICCV paper via introducing an effective module to remove the redundant noise during contextual correlation, which only adds up a small amount of computation burden.
• Memory-Constrained Semantic Segmentation for Ultra-High Resolution UAV ImageryIEEE Robotics and Automation Letters (RA-L), 2024 [paper][arxiv][code]: We present a INR-based ultra-high resolution image segmentation framework, which is efficient under memory constraint.
• A deep learning-based stripe self-correction method for stitched microscopic images
  Nature Communications, 2023 [paper][code]