This temporary repo contains 50k videos generated from 5k text prompts, prepared for VideoScore2.
For the convenience of annotation, we categorize 20+ T2V models into 4 prior quality tiers: Poor/Early, Moderate, Good and Perfect/Real. Then generate 10 videos for each prompt (as a group) and make sure the tier proportion in each group is balanced: 1:4:4:1 or 2:3:4:1.
Overall review of prepared 50k videos: https://docs.google.com/spreadsheets/d/173f9wlDOC24WO3NdghXRTrqF6o37xsxWWta4tTK0wuU/edit?usp=sharing