Researchers have introduced a new dataset, LIVE-YouTube Video Cropping (LIVE-YT VC), to address the challenges of adapting landscape videos for mobile consumption. This database, featuring 1800 videos annotated by 90 human subjects, is the largest publicly available resource for subjective video cropping. The team also developed a smoothed version, LIVE-YT VC++, using a temporal filter to refine annotations. They demonstrated the dataset's utility with the SmartVidCrop algorithm and state-of-the-art video grounding models, aiming to establish it as a benchmark for future research. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT Provides a new benchmark dataset for advancing video aspect ratio transformation models.
RANK_REASON Academic paper introducing a new dataset and methodology.