Introduction to image generation
Ғылым және технология
Diffusion models are a family of machine learning models that recently showed promise in the image generation space. Diffusion models draw inspiration from physics, specifically thermodynamics. Watch this video to learn about the theory behind diffusion models and how to train and deploy them on Vertex AI.
Intersted in learning more about GenAI?
Enroll on Google Cloud Skills Boost → goo.gle/3WeYqMu
View the Generative AI Learning path playlist → goo.gle/LearnGenAI
Subscribe to Google Cloud Tech → goo.gle/GoogleCloudTech
Пікірлер: 14
Subscribe to Google Cloud Tech → goo.gle/GoogleCloudTech
Thanks for the crisp explanation. Very helpful
good job on nice summary
hey so I have a question. in the diffusion model we are ultimately reversing the process of adding noise. so wont it retrieve the same previous image ?
@nanelikahya9949
9 ай бұрын
Yeah how does it get a novel image by reversing the noise addiction?
@jakobullmann7586
4 ай бұрын
From a linear algebraic point of view: It should be clear intuitively that “smearing out” is irreversible, but you can also argue mathematically, e.g. with the fact that the graph Laplacian of any undirected graph is a singular matrix (the graph Laplacian is the generator of a random walk). You train these models to undo tiny amounts of noise. The solution is not unique from a linear algebraic point of view, but out of all possible solutions, only few will look realistic, hence the model is forced to learn the real-world distribution of images. But if you start from an extremely noisy image, there are many possible originals that might have produced this noisy version, too much information is lost.
the awesome stuff! thanks so much for sharing it!
will be a next video ?
Alguien conoce de algún video similar en castellano?
❤
why is the whole video script showing on the screen in the first 2 seconds? 😂
Can you translate in hindi or add subtitles....please....?
Will be a good idea to offer image resize service integreted with your CDN instead of this nonsense...
Honestly it was quite hard to understand