MongoDB and Flask backed online annotation tool for valence and arousal.

The tool is released for research purposes only.

Should you use this tool, please cite:

@article{kossaifi_AFEWVA,
    author = {J. Kossaifi and G. Tzimiropoulos and S. Todorovic and M. Pantic},
    journal = {Image and Vision Computing},
    title = {AFEW-VA for valence and arousal estimation in-the-wild, submitted},
    year = {2016},
}

For more information about the tool, please check the wiki.

You need MongoDB installed.

Clone the repository:

git clone https://github.com/JeanKossaifi/affect_annotation
cd affect_annotation

Create a virtual environment, activate it and get the requirements:

conda create --name annotator python
source activate annotator
pip install -r requirements.txt --user

python manage.py init username password -a 1

python manage.py runserver

This will make the app running on port 55555 at localhost:55555/login

python manage.py new_user username password -a 0

python manage.py update_data

More details on administrating the app in the wiki.

To save the db in a file

mongoexport -d annotations -c annotation -o '/data/savefile.json'

(here we are saving the collection annotation from the database annotations)

To import back:

mongoimport -d annotations -c annotations --file '/data/savefile.json'

python manage.py save_annotations /path/to/save/folder annotator_username

The repo contains some demonstration data. To use yours, create one folder per video in the subfolder:

/annotator/static/data

Each folder should contain images in png format.

Optionally, you can provide annotations in json format. If so, in each folder put a file named like the folder and with a json extension. For more detail check this wiki page.

First install tornado:

pip install tornado

Update the SECRET_KEY in ./annotator/__init__.py:

app.config["SECRET_KEY"] = "your_secret_key"

Then use the deploy script (you may want to change the port):

python deploy.py

When you load the /annotate page, the focus is by default on annotate all frames. If you click that button you can annotate valence using the up and down keys or the sliders. The right key saves the current values and goes to the next frame. When you have annotated all frames for valence, the video is rewinded and you can now annotate arousal with the keys. You can then save as 'check' if you want to verify your annotations later, 'to-do' if you want to redo it or 'done' if you are confident with the results.

For more information about the tool, please check the wiki.

Contributions are welcome, feel free to create a pull-request or open a new issue. If you have questions you can contact me at jean [dot] kossaifi+annotator [at] gmail [.com]