Drag Your GAN: Interactive Point-Based Manipulation of Images
vcai.mpi-inf.mpg.de
Drag Your GAN: Interactive Point-Based Manipulation of Images
1–10 of 90 posts
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#2DragGAN consists of two main components including: 1) a feature-based motion supervision that drives the handle point to move towards the target position, and 2) a new point tracking approach that leverages the discriminative GAN features to keep localizing the position of the handle points.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#3As a technology this is tremendously impressive, straight out of SF movie. However I wonder how it will impact our culture, fashion, standards of beauty etc. when more and more artists will be accepting the generated output then creating their own. Just like in case of music the invention of MIDI, synths and sequencers brought new styles but also boring and imagination-numbing standardization.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#4Neat concept. I wonder if something like that can be applied to diffusion models (the new kid on the block that is outshining GANs right now) - especially since the technique doesn't seem to be too dependent on the generative image implementation.
Also, it's interesting that they're submitting to SIGGRAPH - kind of expected this to be in a more ML-ish conference.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#5Hello, post-truth world!
More seriously, I think that digital photos, and especially low-res surveillance camera coverage, will soon be inadmissible in any reasonable court, because tools like this would allow to forge such evidence in very natural-looking ways.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#6Project website mirror https://web.archive.org/web/20230519060439/https://vcai.mpi-...
GitHub (no code yet, only demo GIF) https://github.com/XingangPan/DragGAN
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#7Neat concept. I wonder if something like that can be applied to diffusion models (the new kid on the block that is outshining GANs right now) - especially since the technique doesn't seem to be too dependent on the generative image implementation. Also, it's interesting that they're submitting to SIGGRAPH - kind of expected this to be in a more ML-ish conference.
Probably goes to show where SIGGRAPH is headed.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#8Hello, post-truth world! More seriously, I think that digital photos, and especially low-res surveillance camera coverage, will soon be inadmissible in any reasonable court, because tools like this would allow to forge such evidence in very natural-looking ways.
Seeing how slowly laws move, the cynic says it's: should be inadmissible in any reasonable court, but will continue to be admissible, and will take a major set of incidents for changes to be enacted across many countries.
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#9Hello, post-truth world! More seriously, I think that digital photos, and especially low-res surveillance camera coverage, will soon be inadmissible in any reasonable court, because tools like this would allow to forge such evidence in very natural-looking ways.
There are probably ways to embed cryptographic hashes within images. Any device that creates images from the real world could have secret keys that can be used to validate any image created by said device.
We will still need a centralized party that holds the secret keys for validation through
Re: Drag Your GAN: Interactive Point-Based Manipulation of Images
#10Hello, post-truth world! More seriously, I think that digital photos, and especially low-res surveillance camera coverage, will soon be inadmissible in any reasonable court, because tools like this would allow to forge such evidence in very natural-looking ways.
There are probably ways to embed cryptographic hashes within images. Any device that creates images from the real world could have secret keys that can be used to validate any image created by said device. We will still need a centralized party that holds the secret keys for validation through
What about the recording device itself? Refeed a video/frame/hash back to a security camera and it tells you if it was originally sourced from that specific camera or not.