Discover and read the best of Twitter Threads about #GANs
Most recents (6)
1/ Day 7 of #AllInAI: Generative Adversarial Networks (#GANs)
Yesterday, we explored synthetic data, which led us to GANs - the tech behind creating realistic synthetic data. After learning more, there are several other real world applications that have come from GANs research… twitter.com/i/web/status/1…
Yesterday, we explored synthetic data, which led us to GANs - the tech behind creating realistic synthetic data. After learning more, there are several other real world applications that have come from GANs research… twitter.com/i/web/status/1…
2/ 👊GANs, or Generative Adversarial Networks, are a type of AI model architecture/technique. They consist of two separate neural networks, a Generator and a Discriminator, that are trained together in a process called Adversarial Training. The goal is to generate new, realistic… twitter.com/i/web/status/1…
3/ 🤖✍️ Imagine a GAN model generating realistic human signatures. The Generator creates a random signature, while the Discriminator classifies it as "real" or "fake." They improve over time, and eventually, the Generator creates signatures that look authentic & are hard to… twitter.com/i/web/status/1…
A team of Mila researchers led by Yoshua Bengio, @SashaMTL, @vict0rsch, @alexhdezgcia today launched #ThisClimateDoesNotExist – an AI-driven experience based on empathy, allowing users to visualize the impacts of the #climatecrisis from #GoogleStreetView images.
We invite you to see for yourself on ThisClimateDoesNotExist.com to visualize the impacts of the climate crisis on the places you hold dear.
The site uses #GANs designed by Mila researchers that generate vivid renderings of environmental impacts, namely #floods #wildfires and #smog
[🧠 GAN paper summary 🧠] "Eyes Tell All" <- An interesting short paper on how to detect fake images generated by GANs (at least the current GAN methods like StyleGAN v2!).
They use a simple heuristic: 1/
They use a simple heuristic: 1/
1. Crop the face and then eyes (they used dlib for this)
2. Segment out the pupils -> you get the predicted mask
3. Fit an ellipse -> you get the "GT" mask
4. Find BIoU between the 2) and 3)
2/
2. Segment out the pupils -> you get the predicted mask
3. Fit an ellipse -> you get the "GT" mask
4. Find BIoU between the 2) and 3)
2/
(BIoU is just a generalization of IoU where we focus only on the boundary pixels of the mask instead of the whole mask)
5. If BIoU is "too low" that means that the predicted pupil is deviating from the elliptical shape and thus the eyes i.e. the face is fake/generated!
3/
5. If BIoU is "too low" that means that the predicted pupil is deviating from the elliptical shape and thus the eyes i.e. the face is fake/generated!
3/
Emtremely excited to finally share a side-project I've been working on for the past few months: Neural Synesthesia, visualizing music with #GANs!
So many ideas left to explore, this journey is just getting started..
More HD renders on my new YT channel:
youtube.com/channel/UCu9a5…
So many ideas left to explore, this journey is just getting started..
More HD renders on my new YT channel:
youtube.com/channel/UCu9a5…
In this project I've built on previous work from amazing AI-wizards like @genekogan, @gwern, @MichaelFriese10, @_joelsimon, @jonathanfly, @refikanadol, @quasimondo & others. All custom models trained using @GoogleColab.
Glad to have created something I find beautiful😌
Glad to have created something I find beautiful😌
For those interested in better quality versions, I've also created a @Vimeo channel with all my #digitalart work:
vimeo.com/neuralsynesthe…
vimeo.com/neuralsynesthe…
1/ I have been asked by many people how I managed to be accepted to @MILAMontreal after two years of trying. Here's my story. Last year I reached out to a new prof at MILA and I audited their class.
2/ They were interested, but when I applied, the semi-automatic system from MILA admission rejected me almost instantly because of my GPA. The admission completely ignored my 6 publications in child #psychology, 2 of which are #statistics paper where I'm first author.
3/ It frustrated me and it still does that this wasn't considered as relevant. Research is research, and research skills are easily transferable from a field to another. People are obsessed with labels and boxes and I didn't tick the correct #AI box.
1/ What are #deepfakes + #AI generated synthetic media; what threats do they pose to #humanrights, #journalism and #publicsphere? What should we fear? What can we do? Thread w/ key tech + usages from recent expert mtg summarizing blog.witness.org/2018/07/deepfa…
2/We've had fake images and CGI for years, so what's different now (beyond enabling us to transplant Nicolas Cage's face to other people). The ease of making #deepfakes is captured by the vlogger Tom Scott
3/ = Barriers to entry to manipulate audio and video are falling - costs less, needs less tech expertise, uses open-source code + cloud computing power. Plus, sophistication of manipulation of social media spaces by bad actors makes it easier to weaponize and to micro-target.
