Computer vision, Surveillance and Camouflage
- Below the threshold value
At the very beginning, Adam Harvey introduced several ways to evade the cameras by wearing makeup and clothing. I really like the idea he presented that you don’t need to change much. You don’t need to be completely unrecognizable. You just need to know what the threshold for computer vision is, and then just need to be lower than that threshold, even just a little bit lower, and you’re invisible to the computer.This brings me back to the excitement mechanism of the human body. It’s also when chemicals break through to a threshold that a burst of potentiation activity (don’t remember the proper noun) occurs. Although mechanistically complex, the key of the decision is still in that threshold and our distance to it. Things seems to be simple while thinking in this way.
- Super low pixel
I was surprised to learn how low the pixel values needed for face recognition and analysis are. Complete face analysis of a person only requires 100×100 pixels of their image. Just like the pattern we use in Gans is 64×64, when I thought 64×64 was not enough, it might have contained exactly enough information. As opposed to a 100×100 photo that defines who we are, i.e. the human perspective. The thing I’m more interested in is the fact that from a machine’s point of view, such a small picture can contain a lot of information.I think the more interesting thing is that Adam Harvey mentions that this can extract the necessary information while protecting privacy. But this is from a human perspective, since a human cannot recognize such a small picture, for example, to see the person’s face in the image. But from the machine’s perspective, the person is already completely exposed under such a single picture. With this in mind, which perspective should we choose for our so-called privacy? The human, or the machine’s? - Samples of database
I remember we learned earlier that machine learning has evolved in this era, a big part of the reason is that this is an era when there is a big explosion of data. Since everyone has enough resources to generate data and disperse data, there is enough data that can be collected. Adam concludes by mentioning that our data is being used by major corporations and governments to train us as machines surveillance.Whether the user is aware of this process of data collection? It reminds me of SketchRNN which collects data from Quick, Draw! game(can’t stop playing for 3 rounds, so much fun). This is a data collection platform that is completely open and transparent to its intention, and at the same time very interesting. I then questioned whether in the future, when data privacy becomes a widely discussed issue, the collection of data could be separated from the data generated by daily use. That is, the collection of data only happens in this kind of game, while the daily data won’t be collected.
Tattoos Model
Here’s the link.