FLIC Video Human Detection Dataset

FLIC is an image dataset of labeled people in frames from movies, which contains 5003 images collected from mainstream Hollywood movies.

The training images come from the person detector running in 30 movies. The images are manually annotated after acquisition, including 10 upper body joints. In addition, the images have 5 median marks to ensure the robustness of outlier annotations. The publisher will actively reject images where the person is occluded or the image clarity is too low. In addition, it reserves about 1016 images of 20% for testing.

This dataset was released by the GRASP Lab at the School of Engineering and Applied Science at the University of Pennsylvania in 2013. The related paper is “MODEC: Multimodal Decomposable Models for Human Pose Estimation”.