Job for Gazelle <https://gaz.ai/>: *Mark "interesting points" red in your logo.*
(corners, edges etc.)

Well. What did Gazelle say?

When asked for a single point, it returned this:

https://botcompany.de/images/1103001

(I could reason about this all day. What an amazing choice!)

When asked for multiple points, it returned this:

https://botcompany.de/images/1103002

_Explanation_

*Top left*: Normal recognition at medium detail level

*Top right*: Recognition at full detail level (almost the complete fur!). This 
is actually too much also, we want something like on the left but more precise.

*Bottom left*: Showing "aura", or "ghost" points (normally filtered out)

*Bottom right*: The grid used in the initial lower detail stages. It doesn't 
actually seem to cover everything, but the detail recognizers appear to recover 
from this just fine.

The whole process is probabilistic, so you quit it early to get rough results 
or keep it running to get more points and more precision.

NB - the algorithm has parameters, but not that many. The main one is "feature 
size" which tells the system how deeply to zoom into every part of the image 
(rounded to a power of two, so it's okay to be very rough there). This will be 
guessed in future versions by simple heuristics & taking  semantic feedback 
from the subsequent processing step into account.

Current recognition time: 20 ms for the medium set of 241 points found in 1,000 
probabilistic steps which are my (relatively) atomic work units.

Current code size: 400 lines <https://code.botcompany.de/1032199> plus libraries
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T2771ec3238f217fa-Mb68235d3339b1ce8294e87f1
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to