We have all read that one article which says that AI is going to completely transform the world, these articles are becoming increasingly common now. Most common examples when people interact with AI are while using Google Search, Beautify Faces, Snapchat filters or Gmail autocomplete. Google Trends show a rather different view where it showed high interest in AI from 2004 to 2008 but dipped between 2008 to 2013. Around 2013, everyone started talking about AI.
But if we start focusing whether its hype or reality, based on the demos of AI which I have seen, AI is already here. Not something in a computer lab from Google but real-world which you can use today. A recent project from Generated Media led to this blog post.
1. AI Generated Faces
1,00,000 faces generated by AI. Where can you use it ? All types of projects -> In your decks, sample websites, presentations, mockups & mobile apps. The global stock photo & video market size is estimated to be $4B. This represents a huge opportunity for entrepreneurs and everyone. Just looking at a sample face makes it difficult for a normal person whether this is generated by AI or a real person. Don’t believe me ? Browse through their Google Drive .
Forecast – Stock Photo on Demand.
Seeing the rapid speed of development, the day is not far where you will only describe the stock photo you require to generate it.
2. Remove Objects in a Video
Not a new demo you must have seen, as there are plenty of apps available already which remove objects from photos like Touch Retouch or Adobe Photoshop Fix but the github project sure looks interesting as it can remove objects from videos. Just draw a bounding box over the object and it gets magically removed. The entire source code is available on Github.
Forecast – Standard feature in all video editors
3. Deep Style Phototransfer
One of my favourite project available right now, you can transform entire tone, style of your photo through this project. The nature photographs taken look breathtaking. It even has a python script to run the transformation for your own images. One of the most popular open-source AI repos with close to 9k stars on github.
Forecast – More Filters available on Instagram & Snapchat
4. Google AutoDraw
One of my most used websites – Google Autodraw – Doodle Icon Recognition.
AutoDraw pairs the magic of machine learning with drawings from talented artists to help you draw stuff fast, but it can be used for almost all types of presentations and projects. Not able to find an icon, or forgot the name of what to search for, use Google Draw to quickly doodle something and find the corresponding icon.
Forecast (or Why isn’t this feature there yet) – Available on Photoshop & Powerpoint to quickly get icons from Doodles
5. Deoldify – Color Old Images
A Deep Learning based project for colorizing and restoring old images (and video!) – very useful to transform black and white photos. The project does what is says – color black and white pictures and videos.
DeepFakes, although controversial, remains one of the most popular examples available of AI. There are countless videos (some very funny!) available on reddit with some of them looking very real. For those who don’t know about Deepfakes – Replace faces in a video with an actor model or your own choosing. When it was first published it was groundbreaking, machine learning available at the click of a button to all. There are pre-built windows app available now.
One real-world application would be in movies similar to how VFX artists were able to bring Paul Walker back in Furious 7 with the help of his two brothers kept him in 350 scenes.
Forecast: (Might be a reality already) Replacing faces in Hollywood stunt doubles scenes
7. Talk to Transformer
AI not only operates on photos and videos but also can operate on text. GPT-2 is the largest model trained on 40GB of Internet text just to predict the next word. GPT-2 is a large transformer-based language model with 1.5 billion parameters, trained on a dataset. You might have seen this in action in Gmail where it presents you with options to complete the sentence. Talk to transformer is an online webapp that creates small stories based upon the a given string of words. The following input of text :
I went to a lounge to celebrate my birthday and
gave me a result
I went to a lounge to celebrate my birthday and I came across some girls who were doing all the karaoke of that night. I thought, hey that's weird," he says. There is another strange thing about the relationship. "I think they're more than friends," he adds, but they still hang out when he visits the hospital for a few days where he works. "It's a bit of a weird relationship," says Kim. "I think they're just like friends. We're friends with our partners. We like to have fun together." But not everyone likes it. "I hate it because I think it's weird. What is a two-hour karaoke session?" asks Alex. But it has a different effect on each partner. Mum is an artist and also likes to karaoke and hang out. And Dad is an IT worker and sometimes karaoke too, but doesn't really enjoy it because he likes to enjoy his home. They are best friends, but he says they are just friends. If they get into something
The above sentence, although not very coherent but is still correct in terms of grammar and has some amount of small coherence built in. A reason might be due to the model in talktotransformer is trained on a small dataset of 700MB, but you can see the rapid advance possible if trained on the real model of 40GB.
Forecast: (Reality already?) You might already be reading news & articles written by AI without realizing it
8. Deep Voice Conversion
Text to speech API are already available from Microsoft, Google and AWS which can simulate real human voice. But deep voice conversion takes it a step further where you can covert the tone, modulation and style of your voice to someone else’s voice. They implemented a deep neural networks using more than 2 hours of audio book sentences read by Kate Winslet. You can listen to both the audios above and see for yourself the potential of the technology.
Forecast: Real audio and AI Audio will be indistinguishable
Have I missed out any project ? Do reach out to me over twitter or comment below