shulie / ppip-dataset

Protect Privacy of Irrelevant People Dataset

PPIP-Dataset

Protect Privacy of Irrelevant People Dataset.

PPIP is a dataset to better evaluate irrelevant individuals recognition in videos, which includes 64 videos, encompassing diverse ethnicities and scenarios. The dataset not only provides annotated coordinates for 141,932 faces in 23,288 frames but also considers the relevance of individuals to the video. It annotates 101,685 irrelevant individuals and 40,247 relevant individuals, facilitating further research in this field.

Due to copyright constraints, we are unable to provide the videos directly. However, we do furnish the specific web addresses of the videos, along with the designated time segments. Concurrently, we offer the result of individuals detection within each video segment, accompanied by relevance labels.

Detail Description

Our data annotation refers to the MOT20 dataset, with the specific format as follows:

[frame], [id], [bounding_box_left], [bounding_box_top], [bounding_box_width], [bounding_box_height], [3D_x], [3D_y], [3D_z], [relevance_label]

Relevance label equals 1 for relevant personnel, and 0 for irrelevant personnel.

Citation

Please cite the following paper if you use our dataset.

Automated Irrelevant Individuals Recognition Algorithm in Video via Motion Trajectories.