The video is great to lead me understand the diffusion paper and Dalle2 !! Thank you Steve!!! Just a quick note, 2:18 in the video, the mapping of random images to raspberry image is not necessary 1-to-1 mapping. For DDPM, the random noise maps to the entire data distribution, which means if you sample one noise image, it could be decoded to different raspberry through different random trajectories.
Re inappropriate images: I'm surprised that nobody's made a version that generates nothing *but* inappropriate images, trained on the huge amount of existing Internet porn. Imagine the transfer capability of DALL-E applied to
You are so good at explaining all of this!
dude your videos are insane quality thank you for making them
The video is great to lead me understand the diffusion paper and Dalle2 !! Thank you Steve!!! Just a quick note, 2:18 in the video, the mapping of random images to raspberry image is not necessary 1-to-1 mapping. For DDPM, the random noise maps to the entire data distribution, which means if you sample one noise image, it could be decoded to different raspberry through different random trajectories.
Great videos, don’t stop making them please
Great explaination!
This is awesome, thank you :)
wow
Re inappropriate images: I'm surprised that nobody's made a version that generates nothing *but* inappropriate images, trained on the huge amount of existing Internet porn. Imagine the transfer capability of DALL-E applied to
Why would you need that ?
@@MODEST500 you are too innocent for the internet
thanks for explaining things for noobs like me
For some reason up votes on this video are not counted 😮
Combine all your knowledge to a paid course, i would love to buy it
You really like chocolate guacamole hah?