Share
View previous topicGo downView next topic
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Computational Perception [study realm]

on Sat Apr 22, 2017 1:32 pm
@evil-mashimaro, come study and share your solutions if you dare.
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 1:48 pm
Nyahhhh nobody shall post in this realm, NOBODY~~~ Ohohohoho

_________________
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 2:49 pm
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 3:02 pm
DARK~~~~ DARK IS BACK TOOOO *36* *36* *36*

WHY DIDN'T YOU POST ON THE SECOND COMING? T_T

_________________
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 3:44 pm
Someone, please do the slides on Convolution and Fourier transform and sounds *29*

Attention

  • People pay attention to where their feature maps are simulated the most
  • Experiment on pop-up phenomena by Anne Treisman (1980s) suggest a paradigm on the speed of visual search.
  • Some target pop-up, so search is constant in time. And others don't, so search is linear in time.
  • Simulation peaks in a feature map or a combination of feature maps leads to pop-up.
  • Missing features and uncertainties leads to linear search.
  • If target is present, the search is on average N/2 (where N is the number of distractors).
  • A computational model for pop-up is saliency map by Koch and Ullman, 1985.
  • They compute a bunch of feature maps and take the most responsive region as the place to pay attention to.
  • But saliency map shouldn't be computed everywhere in our field of view as we have varying acuity. Why? Because it is computationally expensive, and we are only doing one task at a time, which local acuity is sufficient.
  • But this attention thing is different from eye movement like saccade.
  • Two types of attention shift: overt (with eye movement), covert (without eye movement).
  • Posner’s cueing method is designed to study the two shifts and forms the cuing paradigm saying that attention to something increases our performance in that thing.
  • Results are that covert shift is must faster and when attention moves from fovea to periphery, the attention window travels the intermediate regions.
  • In other words (or mine), attention is like another eye.
  • Other people say that attention is like shining a spotlight on a picture.
  • Now people know how to trick others with change blindness and inattention blindness.
    https://www.youtube.com/watch?v=bh_9XFzbWV8
    https://www.youtube.com/watch?v=vJG698U2Mvo


Perceptual Organization, Object Recognition

  • We can group image regions that are similar in proximity, good continuity, color, size, orientation, motion, etc. or shapes that are closed and convex (high in prior probability).
  • Uniform connectedness is a term made to describe regions in image that have some feature that's uniformly distributed, thus we perceive as one image region.
  • Elder and Goldberg develop Bayesian model for contour grouping (2003), considering good continuation, proximity and illuminance similarity cues.
  • In 3D, we also consider reflectance (type of material) and illuminance (direction of light) for edge grouping, and distinguishing occlusion and transparency.
  • The classical object recognition model in Marr’s theory: input 2D image -> detect lines, edges, textures, shadings (2D) -> detect local surfaces, slants, tilts, depth (2.5D) -> detect volumes and surfaces (3D) -> match to object model in memory.
  • This is hacky, don't how to map 2D to 2.5D, then from 2.5D to 3D, and we can't get accurate result in the first 2D step.
  • Now, we use CNN.
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 3:49 pm
You sent an email and asking us to post things again.

I did want you wanted.


Last edited by Dark on Sat Apr 22, 2017 3:52 pm; edited 1 time in total
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 3:50 pm
Okay, this is ma land too *33*

LET'S POST THINGS~~~~ mm1

_________________
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:29 pm
I'm having a hard time with convolution... can someone give a tutorial?
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:30 pm
Which part are you having trouble with?
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:31 pm
... everything ^^"


Last edited by sweet. on Sat Apr 22, 2017 7:34 pm; edited 2 times in total
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:31 pm
Yo, that's too much to explain...
Ask the rabbit?
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:34 pm
Like what's the difference between convolution and cross-correlation?
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:39 pm
Not much. Cross-correlation is like taking a template, centering at each position in the picture and do inner product. Convolution flips the template along all coordinate and do the same thing. But convolution is interpreted as the response of the impulse functions present in the picture, that's why we often use Fourier transform which writes the image with its basic building blocks (the sine and cos functions).
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:44 pm
You mean computationally, they aren't that different. But they have different meaning (just like the ideas of a transcendental force appears in all cultures. The ideas are in essence not that different, but human interpretations, mehehe... spawn a chaotic diversity. mm20 )

_________________
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:47 pm
Haha, yeah it's like seeing a template from two different sides. Very Happy
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:48 pm
YES~
You see my @sweet., convolution and cross-correlation is like the bible and the quran ohohoho

_________________
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:49 pm
o_O why is that even a valid comparison?
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 7:50 pm
Because the rabbit says so. Twisted Evil
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 8:22 pm
=____=

So what's the relation of Fourier transform with convolution?
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 8:32 pm
it's like matter and spirit, li and qi ~~~ mm24 mm24 mm24

_________________
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 8:33 pm
囧 I'll believe you if you write that in the exam and get full points.
avatar
Messages : 108
Date d'inscription : 2007-10-25
View user profile

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 9:58 pm
I think FFT doesn't have anything to do in particular with convolution. Dark is messing you up. FFT just transforms images or sounds from spacial domain to frequency domain. There's just a theorem that links FFT and convolution together.
avatar
Admin
Messages : 605
Date d'inscription : 2007-10-20
View user profilehttp://secondlife.1fr1.net

Re: Computational Perception [study realm]

on Sat Apr 22, 2017 10:06 pm
Yes! the theorem says that THE MEANING OF THE THINGS MADE WITH QI ARE THE QI OF THE MEANINGS OF ALL THINGS~~~~ mm20

_________________
avatar
Messages : 138
Date d'inscription : 2007-11-03
View user profilehttp://vien-ici.1fr1.net/

Re: Computational Perception [study realm]

on Sun Apr 23, 2017 10:31 am
囧囧囧囧囧囧囧囧囧囧囧囧囧囧囧囧囧囧
evil-mashimaro, you're stressing me even more

Thanks, JellyFish Smile and welcome back!
avatar
Messages : 232
Date d'inscription : 2007-11-17
View user profile

Re: Computational Perception [study realm]

on Sun Apr 23, 2017 10:42 am
LOLOLOLOL

sweet, you can also use FFT and apply convolution theorem to find the convolution of a function without actually computing the convolution.
Sponsored content

Re: Computational Perception [study realm]

View previous topicBack to topView next topic
Permissions in this forum:
You cannot reply to topics in this forum