Researchers have introduced HVEval+, a large dataset for assessing the quality of AI-generated human-centric videos, building upon the previous HVEval dataset with pairwise preference annotations. This new dataset includes 1,000 prompts, videos from 24 text-to-video models, and extensive human annotations covering spatial quality, temporal quality, and text-video correspondence. Alongside the dataset, the team proposes MoE-Rater, a multimodal large language model-based method designed for multi-dimensional quality rating, comparison, and question answering within a single framework. AI
IMPACT This work provides tools for better evaluating and optimizing text-to-video models, potentially accelerating improvements in AI-generated video quality.
RANK_REASON The cluster describes a new academic paper introducing a dataset and a model for AI-generated video quality assessment. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
