Protestor-Friendly Live Feed

Video feed of a face being blurred
A filtered applied to a bounding box in a dataset

Inspiration

We drew inspiration from the wide range of political movements that occurred in the past few years, such as the Hong Kong protests, Black Lives Matter, the Belarusian protest, and others, as well as the repercussions that the protestors faced for marching against injustice.

What it does

This product allows the automatic blurring of all faces in a video feed with Gaussian blur.

How we built it

We used opencv for image processing, Google Vision API for all image and face detection, and PyTorch to build the generator for the GAN.

Challenges we ran into

Due to the large nature of computer vision datasets and the need for extensive convolutional neural networks, training the DCGAN took too long, so the end result was that the GAN was unable to showcase its capabilities.

Accomplishments that we're proud of

We feel that theoretical construction of the DCGAN is robust, and Google Vision works very in detecting faces consistently.

What's next for Protestor-Friendly Live Feed

Next is using image segmentation to refine the boundaries and blurring, as well as further training the blurring GAN so that it can totally obfuscate someone's face and be impervious to image sharpening AIs.