>>10597815 (pb)
80% threshold for positive match. They will get a lot of false positives set that low, but that is the intent. Ok.
This implies that if 20% of the image is completely different (in every pixel), then it should pass.
think of it like this, if 7 of the 32 distinctive sub images do not match, then it passes.
think collages.
you only need one,
Soon it will have to be done by rotating 7 out of 32 sub images. If we do collages it should work better imho, because you can just swap out 25% of the images.
This will only work for a few days.
After which we will probably need to be rotating 25% of each image we reuse.
AI can still match this.
Think of it this way. If 7 (6.4) of the 32 sub-images of this image do not match, then it will fail. Blurring, rescaling, jpeg noise is already taken into account.
How does a private political organization obtain "cyber-weapons" developed for the military legally.
For those who don't yet understand what the substructure's of the image are that need to be 20% different every time.
Consider this image.
"High level features" are made from mid-level features, which are made from low level features.
If 20% of the features do not match, then it is a different image. Consider from the mid-level feature level.
AI solved this years ago for assembly line work (because items arrive at different angles all the time)