Ground truth is real data that’s been verified to be correct. Eg a bunch a images that are labeled “hot dog” or “not hot dog”. This is a common AI industry term.
Structured data just means that it’s in a consistent machine-readable format that programs can easily interact with. This is just a common computer science term.
Guided means that it’s not operating from nothing, the process is being influenced or constrained by something.
Taken together it means that they’re taking input data (probably the existing artwork), converting it into a structured data format that’s easier to work with, then that’s used to augment or restrict the AI process to move in the right direction or not go off the rails.
I say it’s misleading because this is something that was fairly obvious to everyone looking at it, even if they didn’t know that terminology. What their model is producing is obviously influenced by and recognizable as a modified version of the original. Even if the overall outcome is lacking.
I think they’re saying that it’s not generating slop from nothing. They take the artist’s “structure data” as a “ground truth” and the generation is “guided” to generate slop that won’t deviate too far from the original?
I know what it means, but it is using jargon to say something simple to mislead people
What does it mean? To me it’s like words pulled randomly out of a hat. (i.e. the emperor has no clothes on.)
Ground truth is real data that’s been verified to be correct. Eg a bunch a images that are labeled “hot dog” or “not hot dog”. This is a common AI industry term.
Structured data just means that it’s in a consistent machine-readable format that programs can easily interact with. This is just a common computer science term.
Guided means that it’s not operating from nothing, the process is being influenced or constrained by something.
Taken together it means that they’re taking input data (probably the existing artwork), converting it into a structured data format that’s easier to work with, then that’s used to augment or restrict the AI process to move in the right direction or not go off the rails.
I say it’s misleading because this is something that was fairly obvious to everyone looking at it, even if they didn’t know that terminology. What their model is producing is obviously influenced by and recognizable as a modified version of the original. Even if the overall outcome is lacking.
I think they’re saying that it’s not generating slop from nothing. They take the artist’s “structure data” as a “ground truth” and the generation is “guided” to generate slop that won’t deviate too far from the original?