The goal of this project is to find conversational groups among people in the image as well as the group membership. More specificlly, we want to predict F-formation, which is a spatial arrangement of people's social behavior.
To predict F-formation with estimated positions and orientations:
- Multi-camera Multi-target Detection: https://github.com/pierrebaque/DeepOcclusion
- Tracking (Optional): https://github.com/cvlab-epfl/pyKSP
- Orientation Estimation and F-formation Prediction: Run GroupVoting_estInputs.ipynb.
To predict F-formation with annotated positions and orientations: Directly run GroupVoting.ipynb.
SALSA Dataset: http://tev.fbk.eu/salsa
Using GT positions and orientations
Using estimated positions and orientations