YT YouTube2Text

YouTube2Text → Transcripts → Perception: 4.4 Object Recognition

Perception: 4.4 Object Recognition — Transcript

by Dr. Kristin Atchison · 2,722 words · 402 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Dr asherson here and we are still on
0:03chapter four recognizing visual option
0:05objects um we're on the home struct so
0:08this is our last video uh video four
0:11object recognition and top- down
0:14processing so for object recognition um
0:17the visual system May recognize an
0:19object as being the same despite the
0:22fact that the retinal image may be
0:24different um and this is called
0:26invariance and there's two approaches to
0:27optic recognition um the first is that
0:31there's a single representation that is
0:33activated when we see an object um and
0:36that all of these objects are show
0:38represented the same way okay so you see
0:41a cup um it's represented there's one
0:45representation for all of the cups um
0:48and every time you see a cup it's
0:50represented the exact same
0:52way the other approach to object
0:55recognition is that object rep are
0:57represented in a view specific manner um
1:00so we can have as many different
1:02representations as there are ways of
1:05seeing them so think back to our
1:06Brooklyn Bridge picture um when we had
1:09all those different views of the
1:11Brooklyn Bridge the variable views that
1:14those would be presented those are be
1:16represented in every specific way that
1:20it could be
1:22viewed so we have a Launchpad exercise
1:26um that will walk us through shape
1:27representation um in area V4 and Beyond
1:32um and so it talks about information in
1:34V4 it talks about the it um cortex and
1:38it will walk you through kind of this
1:40hierarchy um about complexity of stimuli
1:43so do that again for
1:46great okay so in object recognition um
1:50one of the ideas is that there's
1:52something called a grandmother cell and
1:54the idea is um the way it got its name
1:57is that this cell would would respond to
2:01your grandmother's face when your brain
2:03would respond to your grandmother's face
2:05this neuron in your brain would respond
2:06to your grandmother's face
2:09regardless of kind of how your
2:12grandmother's face is presented if you
2:13see a picture if you just hear her name
2:15if you see her face to face if you see
2:18her from far if you see her close up
2:20that this one um neuron would
2:24fire it's this idea that cells are so
2:27tuned that they can respond to a
2:29particular object the problem with that
2:31is think of how many objects there are
2:34out there for us to represent in our
2:37cortex um there's too many cells and it
2:39would require um to be able to recognize
2:43the huge number of different
2:45recognizable objects so there's a back
2:48and forth of do grandmother cells really
2:50exist um this idea was first brought
2:53forth in the'
2:5460s um but there's been some more recent
2:57support for it as well um so they were
3:00studying um epilepsy patients and they
3:03were doing single neuron recordings um
3:06like we've talked about before where
3:07there's one neuron being recorded and
3:09again we were only able to do these
3:11kinds of um studies um because they were
3:14already in their brain so they weren't
3:16going into their brain specifically to
3:18do single cell recordings um but this is
3:20with hey we've got your brain open can
3:22we just chust this um so they were
3:24testing epilepsy patients um and when
3:28they were either they found um in one
3:31situation a neuron responded to a
3:33picture of the Sydney Opera House and
3:35that exact same neuron responded to the
3:38words Sydney Opera house um which really
3:41supported the idea that there may be
3:43grandfather grandmother cells um and
3:45this was done much more recently in 2005
3:49um so it's an ongoing Theory um that's
3:52being tested um and that has some
3:55support and it may be that we have these
3:57grandmother cells for certain kinds of
3:59objects um but not
4:02others so the conclusion is um that
4:05really neurons throughout the visual
4:07hierarchy contribute to object
4:09recognition um that there may be some
4:12specific cells that are involved um like
4:15we saw with the Sydney Opera house but
4:16it really is not just one cell that's
4:19responsible for the whole object
4:21recognition that it really is neurons
4:22throughout this hierarchy that are
4:26involved and that neurons in V1 and V2
4:29are really providing really detailed
4:32information about precise locations of
4:35edges whereas the neurons in V4 are
4:38giving us information about curvature
4:39and orientation and Contours which is
4:42all stuff that you went through in that
4:44Launchpad
4:46exercise and the neurons in the it
4:48cortex are giving us that information
4:50about complex shape um and those highest
4:54level neurons will give us this um
4:57representation of the concept and that
4:59representation is that idea that it's
5:01not changing it's that same
5:03representation it's invariant it's not
5:05changing regardless um of the
5:08presentation of that that that
5:10representation will be the same so it's
5:12again this hierarchy that V1 and V2 are
5:15giving us information about edges V4 is
5:18giving us information about curvature
5:20and orientation and then those it um
5:23cortex is giving us information about
5:26these
5:27complex um
5:30uh
5:31representations so there's various ways
5:34in which we think that these
5:35representations um are being done and
5:37again there's a Launchpad exercise that
5:39will help you walk through this um it's
5:41specifically talking about um the
5:43difference between modular coding versus
5:45distributed coding so modular coding is
5:48the idea that these things are really
5:50only being represented in these
5:52specialized um brain areas that we
5:55talked about like the fusiform face area
5:58um and the parah hiip aample Place area
6:01um and these specific areas are where
6:04the bulk of the work's being done
6:06whereas distributed coding is saying
6:09that this representation is happening
6:10across many regions of the brain and
6:13hopefully what you get from the loung
6:16pad exercise is that both are really
6:18involved um that the visual system is
6:20representing objects in both modular
6:22coding and distributed coding so again
6:25um we're going to walk you through this
6:27through the Launchpad exercise um it's
6:29really going to focus on the fusion form
6:31face area the parahippocampal place area
6:35um and the extra strip body area um so
6:39those are the three areas that the uh
6:41Launchpad exercise will focus on um to
6:43kind of work you through this idea of
6:46both modularity in coding and
6:48distribution in
6:51coding so we do know that there is a
6:54phase selective region and we'll talk
6:56you'll walk we've talked about that in
6:58the previous chapter in chapter 3 when
7:00we talked about the fusor face area
7:01we're showing you that again um in that
7:04Launchpad exercise and here's a study in
7:07fmri experiment um that used um 30
7:10second sequences of faces versus other
7:13types of objects um here we have houses
7:16objects and or scrambled faces um and
7:19they were shown to these human subjects
7:22and what we saw is that and it was faces
7:25versus objects faces versus houses and
7:28faces versus intact
7:29um intact faces versus scrambled faces
7:32so a scrambled face has all the features
7:34of the face um but they're just not in a
7:36fa like pattern um and what we typically
7:38see is that we don't get that same level
7:41of response um because they're not
7:43seeing his face like so it's kind of the
7:46sum of the parts is not the same the
7:48whole is not equal to the sum of its
7:50parts um is that idea of what we're
7:52looking at scrambled faces versus intact
7:55faces and in all of these the fusiform
7:59dyus was more activated when they were f
8:02um viewing faces um and again this is
8:04where they got the term the fusiform
8:06face area and these images here the area
8:09that's marked in green is that it cortex
8:11that fusiform gyrus um and again you see
8:14more activation with faces um than they
8:18did with either objects or houses or
8:21even scrambled faces so there is this
8:24idea that modularity is going on and
8:27there are particular areas of the brain
8:28that really are focus on particular
8:30types of
8:33images because of this because of this
8:35idea of modularity because we have
8:38particular parts of the brain that are
8:40specific to different types of
8:42representations um this creates a
8:44situation where we can have um damage to
8:48that specific module and we can develop
8:51a kind of visual agnosia um there's
8:54different varieties um visual form
8:56agnosia is the ability um where we lose
8:59the ability to perceive shapes um and
9:01this is again usually from damage to
9:04relatively early parts of the ventral
9:06pathway so all of these are along the
9:08ventral pathway um but this is going to
9:10be earlier on in the ventral pathway
9:14this is not damaged to um the par
9:17parahippocampal Place area this is not
9:18damaged to the fusiform phase area this
9:20is earlier on in that ventral pathway
9:24that what pathway and there's a video
9:27that I've assigned for you to watch on
9:29visual form agnosia prosp agnosia um is
9:32again that recognizing faces and I know
9:35chapter 3 covered it some more um and
9:37again um there was a short video that
9:40talked a little bit about it um as well
9:43but it again is damaged to that fusiform
9:45face area and we see um then an
9:48inability to recognize faces um this is
9:51also highly heritable um we can inherit
9:54prosia as well um it's not just brain
9:58damage and then topographic agnosia um
10:01this is going to be damaged to that
10:02parle parahippocampal place area um and
10:07again this is going to impair us one's
10:09ability to recognize buildings and
10:10spatial layouts and other kinds of
10:12spaces other kinds of places um and so
10:15again this is more evidence for this
10:18idea of modularity when one area is
10:21damaged we see impaired functioning to
10:24that area um again this is that idea of
10:27dissociation and double dissociation um
10:30as well that we talked about earlier in
10:32the chapters um so this is involved um
10:36in this modularity and in these
10:38diagnoses of visual agnosia so this idea
10:41that we have these different modules are
10:43not just coming um from brain studies
10:46where we've looked at activation it's
10:47also coming from disordered studies uh
10:50where we look at damage and we look at
10:52see what kind of deficits occur um
10:54through both dissociation and double
10:58dissociation so up until this point
10:59we've really talked about bottomup
11:02processing um in chapters 2 and three we
11:05talked about literally from the bottom
11:07up right we talked about the spectrum of
11:10light we talked about neurons we talked
11:13about how that information the brain
11:15Pathways that gets into we talked about
11:17the the anatomy of the eye we talked
11:19about modularity in chapter 3 even in
11:22chapter four we've been focused on
11:24bottom up we've taken how do we organize
11:27these things how do we figure out what's
11:28the figure versus the ground how do we
11:30then group these things and then how do
11:32we recognize these things we've built it
11:34from the bottom up but the context
11:39influences perception too take this
11:41image per se um both the when you look
11:45at this you read the cat okay but what
11:48you when you really look at it the H in
11:51air quotes and the A and air quotes are
11:54really the same
11:55image but the context of those the image
12:00of that kind of truncated a or that
12:03angled H however you choose to view it
12:06between the E and the T makes it look
12:08like an H whereas between the C and the
12:11T it makes it look like an A so again
12:14that context really is influencing that
12:16process of perception and that's really
12:19indicative that both bottomup processing
12:22is going but also top down processing is
12:25going too so top down is that opposite
12:27direction um instead of building it from
12:30Sensation up it's going to be thinking
12:33down um it's going to be taking what we
12:37know our memory these sorts of things
12:39and going down when you did that
12:40Launchpad example um
12:44about um the fact of
12:47meaningfulness um on recognizing fing
12:50figure to ground um ambiguity that was
12:53top down processing taking a role
12:56meaningfulness was saying that means
12:58something that's top down processing um
13:01top- down processing is really
13:02emphasizing the goals and the atttention
13:04and the knowledge and the expectations
13:08on perception so our knowledge of how to
13:11read was that topdown information our
13:15expectations then that that shape wasn't
13:18H and that shape wasn't a even though
13:20they were the exact same shape that
13:22context and our knowledge of reading
13:24changed our perception of
13:27them it's not just top down or bottom up
13:30both of those things are working
13:32together and we think that those things
13:34meet up in the vental pathway and this
13:36really is helpful to us because it
13:38speeds up the process um it's kind of
13:41meat in the middle instead of having to
13:43build the entire information from the
13:45bottom up or build the entire
13:47information stream from the top down
13:49we'll go both directions and we'll meet
13:50in the mle and we can speed this process
13:53up and we can recognize objects in a
13:55scene
13:56quicker the brain is always actively
13:59predicting the future what it's going to
14:02see what it's going to need to see and
14:03this is very adaptive for us um this is
14:06very um has a a definite evolutionary
14:09benefit in our past um and it uses both
14:12the information from our recent past and
14:15the information from the current
14:16information that's around us and
14:19unconsciously we make an inference about
14:21what's going on we're guessing kind of
14:25based on our past and based on the
14:28current situation ation and taking those
14:30two things and meeting in the middle
14:31again on this top down
14:35processing theories of top down
14:37processing are really more conceptually
14:39driven um while bottom up processing
14:41theories are data driven so this is why
14:43we've spent so much time talking about
14:46bottom up theories is because those are
14:48data driven and we have a lot of data to
14:50back those up those are really easy to
14:52quantify top- down theories are a lot
14:55harder to quantify which means they're a
14:57lot harder to measure which means
14:59they're a lot harder to do um so we have
15:02a lot um a lot less information about
15:06that we do think that one of the things
15:08that could be going on is the visual
15:10system is taking statistics about our
15:13world okay it's taking both again the
15:15statistics of our past and it's taking
15:18the statistics of our current visual
15:20information and combining those two
15:23probabilities together
15:25unconsciously to kind of guess ahead
15:29um about what's happening in the current
15:31visual field in the current retinal
15:33image so again it's combining the idea
15:36of this prior probability and it's the
15:39prior probability of all possible scenes
15:42okay and so in the image here that we
15:45talked about earlier on in the chapter
15:47that was referenced earlier in the
15:48chapter B and C are both possibilities
15:52for a there's lots of different
15:54possibilities for a but B and C are two
15:57of the many possibilities and what your
16:00visual system is unconsciously doing the
16:02theory says is combining all of those
16:06it's also taking information on the POS
16:09for each possible scene the probability
16:12that that produce the current no image
16:15so in this example B is more probable um
16:19than C is because we have that we have
16:23continuity um for the line the Rope on
16:26the left where we don't have continuity
16:28for the line on the right so that's
16:30going to make it more probable that the
16:33image um what's happening with the ropes
16:36under that fallen tree is B rather than
16:40C but again we're taking both of those
16:43things and making simultaneous judgments
16:45about this um so it's again the idea
16:48that we not only are processing this
16:50information from Sensation up to
16:52perception but we're also taking our
16:53knowledge and our experience and we're
16:55using that information to meet in the
16:57middle and make judgments about
17:00situations like this thanks and I will
17:03see you next time

About this transcript

This page contains the full transcript of Perception: 4.4 Object Recognition by Dr. Kristin Atchison, generated from the public captions YouTube serves with the video. The transcript has 2,722 words across 402 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.