Dataset · litforth
Video group activity and human pose annotation
litforth/Collective-Activity-Recognition
Video frames with manual bounding box annotations of people, their actions, and pose directions for video analytics systems.
What's inside
Contains images (10K to 100K records) extracted from every tenth frame of video sequences. For each person in a frame, bounding box coordinates, activity class, and pose direction are manually annotated. Data is presented in a text format tied to frame indices and coordinates.
What you can do with it
- Training person detectors and video group action recognition models
- Training body orientation (pose direction) algorithms in surveillance systems
- Size
- 10K–100K rows
- Downloads this month
- 26.6K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("litforth/Collective-Activity-Recognition")