redacted message from SLR's slack channel
Hey guys, we have put a little video together for fun. As you might due to the Japanese 🗾 obscenity law know every video got to be censored (pixelated) before it can be distributed to the public. Creating a censored videos is notoriously hard and tedious process where you need to go thru each frame at the video and make a mask for it, censoring one video might take weeks or even more if there are multiple people in the video. It’s also very expensive. So far everything has been done manually with very little automation.
For past few months me, {name1} and {name2} have been working on Japanese censoring project at SLR. We’ve created a deep learning model that is capable of generating these censored videos on its own. This is fundamentally the same system that are powering modern self-driving cars 🛺 , autonomous vehicles, deepfake videos and other cool technologies.
We have collected a dataset of thousands of annotated screenshots to which we keep adding new data daily.
The model only gets better from here, this fundamentally changes the game, hopefully until to the point where human supervision will be minimized to skimming thru the video.
We are also building an auxiliary MosaicAI app with {name3} to fix frames where model didn’t do a good job, these frames will later be added to training set and improve the model in the future.
I expect that we will have a production model in 2-3 months time. This will help Japanese studios to cut the costs down and get videos faster than ever. The censoring made by a machine is much greater quality compared to manually made videos, each frame is perfectly fitted. First production video should be done in less than a month from here. Planned turnaround time for a video should be less than 24 hours.
The potential of deep learning in truly amazing, even with current models so many things can be done better or totally automated.
Please check the video subscribe to the channel, hit the bell button 😅 (keep the sound on)
Check a demo here
Next on in the crosshair: solving scripts and making every scene on SLR interactive.
Anyways, this post is purely promotional, if you are an engineer with background in computer vision, deep learning, neural nets, GANs, labeling, data prep and etc, join us at SLR team! We look at dick pics daily.
My email philip at sexlikereal.com
😎😎😎
edit: also looking for data science people, analytics, visualizations and so on
