Memento10k is a dynamic video memory dataset containing 10,000 3-second videos. Each video has more than 90 human annotations, and the split-half consistency of the dataset is 0.73 (the best among video memorability datasets).