Aeye shiber algorithm

  • 7 Replies
  • 3357 Views
*

yotamarker

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1003
  • battle programmer
    • battle programming
Aeye shiber algorithm
« on: June 15, 2017, 09:14:18 pm »
in an image there are all sorts of objects.
how is it decided on which to pay attention to ?

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Trusty Member
  • **********************
  • Colossus
  • *
  • 5865
Re: Aeye shiber algorithm
« Reply #1 on: June 16, 2017, 01:38:13 pm »
Wouldn't that depend on the assigned goal or task?

Unless it were a game to see "how many shapes or animals / objects one can find", it should be set forth early in the process or what is being sought.

Just my $.02
In the world of AI, it's the thought that counts!

*

yotamarker

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1003
  • battle programmer
    • battle programming
Re: Aeye shiber algorithm
« Reply #2 on: June 16, 2017, 01:42:16 pm »
with the eye there are several factors to consider like:
movement
colors
size

and even if it is something the A.I is searching for she still needs to be aware of other objects.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 824
Re: Aeye shiber algorithm
« Reply #3 on: June 16, 2017, 02:22:26 pm »

 For algorithms that need to be trained:
 Unsupervised learning, needs to see a out line of a object in a different location. ether
 in the same picture or different pictures.    like optical flow algorithm.

 Supervised learning learning uses a data set of pictures that are labeled with the object
in the picture.

 In non training  algorithm there is the "brute force matcher". Where you give a 
 program a descriptor table of the object. Then have it search through a data
base or video.


*

yotamarker

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1003
  • battle programmer
    • battle programming
Re: Aeye shiber algorithm
« Reply #4 on: June 25, 2017, 09:07:18 pm »
here is the puzzle :
an image contains x amount of objects
the higher the resolution the longer the process time
you can :
A sacrifice some accuracy of the image contouring and make up for it with actual memory and thus see the whole thing

B look at little pieces of the image of the image meaning see well far away objects, assuming they don't over blur
and have little to no idea as to what exactly you are seeing.

c beef up the processing power which in theory may get better results

please comment if you agree and if possible elaborate.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 824
Re: Aeye shiber algorithm
« Reply #5 on: June 25, 2017, 11:57:08 pm »
 Once a object has been figured out at high detail, It then can be tract in very low
detail mode. Only need to bring it into
high detail when you focus on it. It become your RO!, region of interest. 

 But then need to learn what  a low resolution object will look like. It will be a simpler
outline and more memory.

 When a person enters a room, they look at blurs and can tell what they will transform
into when they are focused on.

 Since low resolution object  and the outline of low resolution object can be confused
with other objects. A AGI uses this a cornerstone of its logic of substitution.



*

Korrelan

  • Trusty Member
  • ***********
  • Eve
  • *
  • 1454
  • Look into my eyes! WOAH!
    • YouTube
Re: Aeye shiber algorithm
« Reply #6 on: June 26, 2017, 12:11:02 pm »
If you research the human visual system you will note that the sensory neurons in the human retina are mapped to the visual cortex in a polar schema.  Humans don’t recognise all the objects in a scene at once, we flick our focus around noting the objects and their relative positions to each other.

I’m sure I’ve mentioned this before but if you switch to a polar vector schema for the eye then you can have the best of all worlds.

https://www.youtube.com/watch?v=nIkevg21AM4

This method spatially compresses the image, giving a large coverage of periphery and yet retaining the high resolution center/ fovea.  The large area covered by this polar lens can be seen in the bottom left of the left hand window. You can now run recognition routines against the small image and translate the vectors back to polar cords for mapping onto the original image if required.

Keep in mind that an AGI moves its cameras so in reality the center of the lens would always be in the center of the video feed window, but for testing purposes its easy to move the fovea around the image.

Its easy to see how objects in the periphery can still be clearly seen, this can be used as a guide for saliency/ attention.  The vector point angles that define the eyes layout can be easily translated in servo actions, training the center of the eye on the object to aid recognition.  If colour blob or shape is detected at the periphery the machine moves its eyes to focus the high resolution center.

 :)
It thunk... therefore it is!...    /    Project Page    /    KorrTecx Website

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *******************
  • Prometheus
  • *
  • 4659
  • First it wiggles, then it is rewarded.
    • Main Project Thread
Re: Aeye shiber algorithm
« Reply #7 on: June 26, 2017, 12:56:04 pm »
I see polar? I don't though......

Does using those small convolved images that have been polarized make recognition BETTER? Or for what certain images and purposes?
Emergent          https://openai.com/blog/

 


OpenAI Speech-to-Speech Reasoning Demo
by ivan.moony (AI News )
Today at 01:31:53 pm
Say good-bye to GPUs...
by MikeB (AI News )
March 23, 2024, 09:23:52 am
Google Bard report
by ivan.moony (AI News )
February 14, 2024, 04:42:23 pm
Elon Musk's xAI Grok Chatbot
by MikeB (AI News )
December 11, 2023, 06:26:33 am
Nvidia Hype
by 8pla.net (AI News )
December 06, 2023, 10:04:52 pm
How will the OpenAI CEO being Fired affect ChatGPT?
by 8pla.net (AI News )
December 06, 2023, 09:54:25 pm
Independent AI sovereignties
by WriterOfMinds (AI News )
November 08, 2023, 04:51:21 am
LLaMA2 Meta's chatbot released
by 8pla.net (AI News )
October 18, 2023, 11:41:21 pm

Users Online

311 Guests, 0 Users

Most Online Today: 346. Most Online Ever: 2369 (November 21, 2020, 04:08:13 pm)

Articles