Community Archive

🧵 View Thread

🧵 Thread (15 tweets)

Placeholder
UltimApe@ultimape• about 7 years ago

Just blur the picture of the user a bit and then run a similarity algorithm. If the reply is from a different account with a similar image, boost the weight/score that it's a scam tweet.

2 1
7/7/2018
Placeholder
UltimApe@ultimape• about 7 years ago
Replying to @ultimape

Hausdorff distance between two images can be calculated on a GPU since 2011.Why isn't this a solved problem?

4 0
7/7/2018
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

OkCupid a blog post on detecting similar images that have scale and cropping applied.https://t.co/F0dvINf3WZ

2 1
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Here is a method to detect basic photo filter manipulation signature, https://t.co/QNEdDmh1OR

2 1
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Nvidia has a thing that can remove watermarks / text and even de-noise an image quickly. Not sure if it's released, but something like that would be good at removing the stuff OkCupid talked about being thorny problems.https://t.co/nNhQ051UGp

2 1
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Disney also shows you can use similar ideas against videos (at least the denoising one).https://t.co/9C9IkEx9Jg

0 0
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

I can't find much on the system YouTube uses, but they seem to have rolled out a similar distortion detection mechanism as part of their Content ID apparatus. I can't find anything directly about it, reports say its now being done with 'Google Brain'?https://t.co/vRQqYMa5wL

0 0
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Noise detection is neathttps://t.co/3jVC9v3FsI

Placeholder
UltimApe@ultimape• about 7 years ago

The first time I heard the concept of Bayesian control theory was from a ted talk about what brains are for. it was literally on applications of noise detection within robots and muscles. https://t.co/CXhmq4CEST

2 1
0 0
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Two Minute Papers covers that Nvidia thing in more detail: https://t.co/PK3L1xoseq

0 0
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Oooh, it has a link to a knock-off implementation using Kerashttps://t.co/DPWMX4m62m

0 0
1/24/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Anyone know of good sources for information on modern perceptual hashing techniques? This paper talks about using them to track meme lineages.https://t.co/8GZkDBShcvApplying the techniques to a corpus might net interesting patterns like this one: https://t.co/xY3gPE1A2Q

1 0
5/28/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

Ah, its seems someone already did what I suspected was possible.https://t.co/bBnDqm7fqU

0 0
5/28/2019
Placeholder
UltimApe@ultimape• over 6 years ago
Replying to @ultimape

What's wild to me is that their approach was published only a month after nvidia's denoising one, basically suggested they either used the same technique, or discovered the same roughly similar solution due to similar constraints.

0 0
5/28/2019
Placeholder
UltimApe@ultimape• over 4 years ago
Replying to @ultimape

Belhassen Bayarhttps://t.co/GzAnwOPZx7 andMatthew C. Stammhttps://t.co/mhdrj1CDSLHave some great stuff on detecting image manipulation. Some of the latest work accounts for nuances in camera make/model.https://t.co/Km12uBBsiIMight be good to detect deep fakes.

0 0
2/10/2021
Placeholder
UltimApe@ultimape• over 2 years ago
Replying to @ultimape

I am excited about algorithms to thwart spoofing of images and fakes.https://t.co/3gczPDFKMZ

Placeholder
UltimApe@ultimape• about 6 years ago

Compression, divergence. Shared meaning and deviation."Pompeiu–Hausdorff distance, measures how far two subsets of a metric space are from each other."https://t.co/PnC2xAVpNohttps://t.co/dqfz8LTfPc

0 0
0 0
5/30/2023