Reproduce “Event-assisted Low-Light Video Object Segmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hebei Li, Jin Wang, Jiahui Yuan, Yue Li, Wenming Weng, Yansong Peng, Yueyi Zhang, Zhiwei Xiong and 1 more
Researchers at the University of Science and Technology of China developed an event-assisted framework for low-light Video Object Segmentation (VOS), introducing the first synthetic and real-world low-light event-VOS datasets. The method features an Adaptive Cross-Modal Fusion module and an Event-Guided Memory Matching module, achieving superior performance on both new datasets.
Key insights
- Event camera data, providing robust motion and edge information, significantly complements degraded frame-based visual input, proving essential for VOS tasks in low-light environments.
- Effective multi-modal fusion in challenging lighting conditions requires adaptive strategies to handle noise in low-light images and selectively integrate precise event data for contour and detail restoration.
- Event-guided mechanisms, combining event and mask priors, can robustly steer memory-based object matching, overcoming the unreliability of purely image-based features in low-light VOS.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.