Pass: An ImageNet replacement for self-supervised pretraining without humans
1–10 of 12 posts
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#2This is amazing, so much value for the researchers
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#3I wonder how much worse the performance will be on human-related downstream tasks.
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#4I wonder if that's a good idea.
One could easily argue that throughout all of human evolution, recognizing and interacting with other humans was the single most important factor for survival. Accordingly, one would expect the human visual system to be highly adapted towards recognizing humans. [1]
If we want AI to safely interact with humans, then we should strive to make the AI understandable in the sense that our intuition of what it might see or do matches up with what it'll actually see or do. That's a big aspect why those Tesla crashes are so scary: No human would overlook a huge white truck sideways on the road. But an AI trained on the wrong dataset might.
Accordingly, I would have expected that datasets move more towards imitating what babies will see in their first days or weeks, because that seems to be the most promising path towards replicating human vision with AI.
[1] https://news.stanford.edu/news/2012/december/infants-process...
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#5I wonder if that's a good idea. One could easily argue that throughout all of human evolution, recognizing and interacting with other humans was the single most important factor for survival. Accordingly, one would expect the human visual system to be highly adapted towards recognizing humans. [1] If we want AI to safely interact with humans, then we should strive to make the AI understandable in the sense that our i…
The car industry's argument against this would be: who cares if the car makes a mistake a human whould never make if the overall statistics of (fatal) accidents go down?
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#6I wonder if that's a good idea. One could easily argue that throughout all of human evolution, recognizing and interacting with other humans was the single most important factor for survival. Accordingly, one would expect the human visual system to be highly adapted towards recognizing humans. [1] If we want AI to safely interact with humans, then we should strive to make the AI understandable in the sense that our i…
The car industry's argument against this would be: who cares if the car makes a mistake a human whould never make if the overall statistics of (fatal) accidents go down?
Also my argument.
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#7Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#8I wonder if that's a good idea. One could easily argue that throughout all of human evolution, recognizing and interacting with other humans was the single most important factor for survival. Accordingly, one would expect the human visual system to be highly adapted towards recognizing humans. [1] If we want AI to safely interact with humans, then we should strive to make the AI understandable in the sense that our i…
They do say the dataset is for pretraining. If your application needs the ability to identify humans, then your fine-tuning dataset would include humans.
This seems totally appropriate for pretraining on natural images IMO.
Re: Pass: An ImageNet replacement for self-supervised pretraining without humans
#9I wonder if that's a good idea. One could easily argue that throughout all of human evolution, recognizing and interacting with other humans was the single most important factor for survival. Accordingly, one would expect the human visual system to be highly adapted towards recognizing humans. [1] If we want AI to safely interact with humans, then we should strive to make the AI understandable in the sense that our i…
They do say the dataset is for pretraining. If your application needs the ability to identify humans, then your fine-tuning dataset would include humans. This seems totally appropriate for pretraining on natural images IMO.
Wouldn't your pretrained model be lacking in neurons that react to human features though? Seems like a poor choice if your ultimate goal is human images.