Check this out:


Gazelle's GUI shows how it sees the world (in this case, a computer screen.

The preprocessing step is as simply as it is (I think) brilliant: We just 
reduce the number of pixels and colors... and convert everything to grayscale. 
Suddenly segmentation gets a whole lot easier.

Put yourself in the AI's shoes. What do you see in the image above?

Gazelle can recognize continguous regions of the same color. We could start 
with that and choose a few regions in the image for further investigation.

Note that we actually don't know yet what the original screen shows. So we are 
in the same position as the AI, which is I think a good way to approach 
problems like this.

(Solution BTW: It's this album cover 
<https://www.youtube.com/watch?v=J_nSXbAPdFA&list=RDR8pjo7yvOJc&index=23&ab_channel=EDIZIONIMONDO>.)

So next I shall add a way to click on a region (they get highlighted already 
when you hover over them) and store it.. optionally with a label.

And then we can start comparing shapes. And stuff.

Latest download (JAR) here: https://code.botcompany.de/1033636 Should run on 
any major OS with Temurin 17 <https://adoptium.net/>. (Apparently they renamed 
Java.)

*BTW I am urgently looking for another small-ish investment (four digits € or 
$). Write to **[email protected]** if you are interested.*

More example of "insect vision" (slightly better than insect vision actually, I 
think insects see in 16p or so):



Makes you want to zoom in to see what it actually is, no? That's exactly how 
Gazelle will work.

Cheers
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T2dea42566ccf6693-Mbc199bb767a66d81047f2169
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to