Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
Divide-By-0 authored Jul 13, 2023
1 parent b302453 commit 3472478
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -242,7 +242,7 @@ Ads bring in surprisingly low revenue -- I hope that the bounty makes it worth y
- Auto Analyze Game Footage: A drone above a sports game keeps track of all the footage and then you infer the plays the opponent is making based on similarity of movement, so you can get the strategies of an arbitrary team. Can also do with phone cameras possibly.
- A hidden camera on a lapel that lets you record all poker hands at a casino and then analyses them for you.
- Trippy Video Generator: A superresolution/style transfer model runs on each frame of a video independently, leading to a very raw flipbook-style animation. Perhaps generate a trippy video from a regular video. For instance, upresolutioning every frame, or recoloring each frame (or any other transformation) independently with an imperfect generative adversarial network to have disconnected images but a connected idea.
- Note: This got built! See https://aimlabs.mit.edu/ > Stylish Videos!
- DONE: This got built! See https://aimlabs.mit.edu/ > Stylish Videos! These days, stable diffusion basically does this pretty well.
- Fill in people's videos when their cameras are off. By using the speech-to-face paper in a browser extension, fill in other's inactive google hangouts/zoom logos with low bitrate moving bitmojis, and send yours as that when your camera is off. Cool research done at 5.3 (https://arxiv.org/pdf/1905.09773.pdf).
- Given a (lecture) video, edit it so that automatically speeds up during times where no actual lecture content is delivered (ex. an instructor talking about their dog), and slow down when content is dense (ex. slide full of equations that the lecturer is going through). Use both visual info (slides) and audio (what the lecturer is saying). Example use case: a student wants to watch a recorded lecture video in an optimal manner that maximizes learning in a shorter amount of time.
- Live transcribe any Chrome tabs audio by pushing it into OpenAI Whisper-type model, enabling transcriptions for all meetings including gather.town out of the box. Can have a start/stop button that auto-saves it to a Google Doc.
Expand Down Expand Up @@ -278,7 +278,7 @@ Ads bring in surprisingly low revenue -- I hope that the bounty makes it worth y
- Keyboard shortcut for automatically fixing spelling in Google Docs+.
- A chrome extension that adds the per ounce/unit price to Instacart the same way Amazon does.
- **Fix Forgot to Reply To Message**: App that finds old messenger chats or emails you left on read or unread, or unfinished threads, so you can respond to them. Requires probably mapping the last few texts to a score of how likely it was to be the end of a conversation.
- Talk to a 'painting' and have it talk back, like Harry Potter paintings. Speech-to-text on your voice, gpt-3-type-model for conversation, tokkingheads/other deepfakes to have a face emulate the response. This has been [attempted](https://github.com/Halcyox/XRAgents) but not yet real-time end to end on arbitrary humans, and nowhere close with animation on cartoons.
- Talk to a 'painting' and have it talk back, like Harry Potter paintings. Speech-to-text on your voice, gpt-3-type-model for conversation, tokkingheads/other deepfakes to have a face emulate the response. This has been [attempted](https://github.com/Halcyox/XRAgents), but not yet real-time end to end on arbitrary humans, and nowhere close with animation on cartoons.
- Create a simple scraper that searches Google for [specific bot websites](https://twitter.com/gregegansf/status/1363978958841155585) and reports/blocks them. This is more a way to get to talk to/help Greg Egan, than it is a real problem.

<a name="Fun"></a>
Expand Down

0 comments on commit 3472478

Please sign in to comment.