Conflab dataset repository

This repository contains code that may be useful for use of the Conflab dataset.

We provide:

• Baselines used in the paper:

  1. video pose estimation and person detection
  2. f-formation detection
  3. speaking status detection from pose and acceleration

• Preprocessing code used to process the raw data.

• Data loaders. We provide the data loaders (torch datasets) used to run our baselines.

• Analysis. Contains some of the code used to produce the visualizations in the paper.

Setup

To use the code externally, add the parent folder to PYTHONPATH. For example:

git clone [email protected]:TUDelft-SPC-Lab/conflab.git
export PYTHONPATH=$PYTHONPATH:$PWD

Then you will be able to import conflab in your own projects. For example, to use our person data loader in your own project:

from conflab.data_loading.person import ConflabPersonDataset

Path to the dataset

The dataloaders and preprocessing scripts need to know the path to the conflab dataset files. You can set it in constants.py:

conflab_path = '/path/to/my/conflab/copy'

Baselines

The speaking status detection baseline is part of this repository (except for the MS-G3D pose-based method which is in a fork of the original author's repository):

• Speaking status detection from pose

The F-formations and person detection / keypoint estimation baselines are in separate repositories:

• F-formation detection
• Keypoint estimation

All these baselines are also referenced as git submodules in this repository under /baselines. To download them, run the following after cloning:

git submodule init
git submodule update

or clone with --recursive option:

git clone --recursive [email protected]:TUDelft-SPC-Lab/conflab.git