Inspired by the brutality of the George Floyd video and the riots this library seeks to give a methodology and machine learning library for automated identification and mechanism for alert of current brutal situations that are being recorded.
Violence as a response in any situation does not move us in the right direction. We will identify all violence and brutality, then only non-violent responses should be looked at as responses to these tough situations.
TODO: Add Medium Article on the subject
Help Wanted - Come help build products that change the world.
v0.1
- Privy Filter - Photo Privacy Library
- Notebook get all Characters and Poses in Scene
- Notebook get all Characters Skeletal Depth Location
- Notebook for dataset generation - Will complete with labels
- Manual data labelling of minimum 1000 pictures
- Train/test labeled dataset of scenes and pipeline of character skeletal location data
- Logistic Regression(May Change) on Vector Skeletal Differentials Classification Model Training
- v0.1 Python Package Release
v0.5
- Review Methodology Accuracy and Corner Cases to Determine new v0.5 Methodology Improvements
- Facial Aggression / Age Detection / Sex Detection - Additional Features using the deepface library
- Roadmap Planning for Next Gen
- Facial Emotion - Identify face and level of aggression to add to the model.
- Age Detection -
- Sex Identification
- Gun Identification
- Blood and Cuts Identification
- Video Support
- Audio Support
Please note this methodology will improve over time and we will use this repo to build out a labeled training and test dataset to increase accuracy.
v0.1
- Ensure the Media is Recent - Identifying brutality in the past may have value but we want to make a library for the actionable real world and there will already be a lot of movies/videos that will trigger the algorithms so we will focus on photos marked within the last 24 hours just in case timezone issues and keep the first version simplistic.
- Compute 3D Human Positions - We combine human pose estimation, and machine learning depth of the image to get a 3D representation of all the humans identified in the pictures.
- Calculate the Likelihood of Brutality - The first version will use a vector posture differential between multiple characters in the scene. If it appears certain characters are close enough and in a dominating position over another
Use the package manager pip to install George Floyd library.
pip install george-floydimport george-floyd
(bruttality, confidence) = george-floyd.process(img)Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update the tests as appropriate.