Researchers have developed a new framework called the Spatial-Temporal Multi-scale Network (STM-Net) designed to enhance the quality of screen content videos (SCVs). Unlike natural videos, SCVs present unique challenges due to abrupt motion, scene changes, and high-frequency details like text and graphics, which can degrade performance in conventional video enhancement methods. STM-Net addresses these issues with three key components: a Prior-Guided Spatio-Temporal Dispatcher for parallel processing, a Bidirectional Temporal Feature Extraction module for handling transitions, and a Cascaded Multi-scale Feature Distillation module to preserve fine details. Experiments show STM-Net surpasses existing methods in both objective and subjective evaluations. AI
IMPACT This research introduces a novel network architecture for improving screen content video quality, potentially benefiting applications requiring clear text and graphics in video.
RANK_REASON The item is a research paper published on arXiv detailing a new network for video enhancement. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Bidirectional Temporal Feature Extraction
- Cascaded Multi-scale Feature Distillation
- Hugging Face
- PG-STD
- Prior-Guided Spatio-Temporal Dispatcher
- Spatial-Temporal Multi-scale Network
- STM-Net
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
