August 09 2023, 15:44

now it’s possible!

https://arxiv.org/pdf/2308.01074.pdf

The guys presented a practical implementation of a deep learning model for classifying laptop keystrokes using a smartphone microphone. When trained on keystrokes recorded by the nearest phone, an accuracy of 95% was achieved in determining the text typed by a person based only on the sounds of the keyboard. When trained on keystrokes recorded via the Zoom application, an accuracy of 93% was achieved. These are quite interesting results.

August 06 2023, 12:07

To my surprise, I found that drawing cars is more difficult than drawing faces. I tried to sketch our RAV4 “quickly” but got the proportions a bit wrong. The challenge is that you draw the car standing next to it, and because of this, the perspective distortions are much stronger than with smaller objects at a greater distance. Additionally, the shape of a car is defined by a continuous array of intricately inclined planes and transitions between them, rarely aligning with the lines of perspective. Plus, specific details distinguish the car against other brands, and I still need to figure out which ones, other than the front of the car. Overall, it’s an interesting exercise that is still challenging, but I will keep trying. Luckily, there are always enough parked cars on the street.

August 05 2023, 00:14

I’m a few years late with videos about robotaxis on city streets, but just in case someone hasn’t seen it yet. There are quite a lot of these little cars. Well, how many… to film them, you need to wait about 15 minutes in the city center – at least one will definitely pass by. Not in the center, it might take a bit longer. Primarily, you see the Cruise (GM) and Waymo (Google) cars. They operate day and night, even during rush hours. You download the app, place an order, and off you go.