Consent of protesters / anonymizing video

[mjmaurer]

The reality is that any centralized cataloging of videos and pictures could also be a tool for law enforcement. Does anyone know of tools that could be used to anonymize video?

Otherwise, any faces in a video could be looked up in a tool like Clearview AI.

[mjmaurer]

Youtube claims the blur it adds to videos cannot be practically reversed: https://youtube-eng.googleblog.com/2017/08/blur-select-faces-with-updated-blur.html

is there a similar service that exists?

[miker2049]

something like opencv through js could achieve this if we wanted to keep it as a webapp/service? I see lots of little tutorials (e.g.), but no existing service. This is something I can try and work on! I am no professional, but got some good years of self taught digital image processing. And this seems as simple as just applying a haar-cascade classifier on each frame to find a face and they process that.

I know there is concern about the ability to unblur faces with not-as-sophisticated blurring, but it should be fairly simple to just have a black square?

This repo just has links (right now) to social media, so I assume this should be more connected to the pb-videos repo?

[zimmertr]

This is a hard problem to solve and I definitely agree that this is an ethical issue that warrants a deeper discussion.

To play the devil's advocate, should one have an expectation for privacy in a public place? To my knowledge, that right is not usually enforced in the United States of America. This is a topic that is heavily debated with respect to body cams and protests. Some people suggest that it is up to the citizen to mask their identity in public if this is a concern. This is a unique situation though...

I understand the risk of this project becoming a honeypot. Would it be fair to assume those who have been brutalized in these videos have already had their identities determined via proxy of viral spread and/or the associated officers? Would the duplicate presence on this project pose an added risk?

Some other people have started similar initiatives that are well beyond this one in public recognition. For example, this lawyer on Twitter who has aggregated over 200 instances in a very popular thread. Has the damage already been done with respect to leaking identities?

I am also inexperienced with image detection, but I suspect using a framework to evaluate videos and automatically blur identities might be quite expensive in terms of compute. It will certainly increase costs for this project by several magnitudes as it will likely be necessary to purchase compute capabilities with GPU support for image processing.

[miker2049]

You bring up some good points.

I think it all kinda of makes me realize that I might need a clarification of the goals and purpose of this specific repo. We are kinda of right now treating the project as a combination of frontend, backend, and content elements, and its the last one that doesn't usually get a lot of play on github.

Further, this effort, if it happened, would need some kind of specification with what happens in #32, and seems to be the more appropriate place to discuss the actual necessity of this?

All that said, opencv.js and face-api.js can do face detection/processing browser-side very well (with pre loaded cascade models), I can't imagine a gpu would be necessary, especially if this is presumably a tool that will just scrape a video from a twitter/fb url and process it before mirroring. Nothing real time needs to happen right?

[mjmaurer]

Has the damage already been done with respect to leaking identities?

I think adding to IPFS will guarantee damange so why not try to limit beforehand? I think the JS in the browser idea!

[00000100-00010100]

Signal has face blur for images now, could be useful: https://signal.org/blog/blur-tools/

[EndingPoliceBrutality]

I'm not sure that police would even be able to use the video in a criminal trial (if that was the original concern). They have no way to demonstrate chain of custody. Also with providing incident details and location the police may be able to just lookup the original source of the video anyways, making the blurring (that is computationally expensive) pointless. Not a bad idea, just some of my thoughts.