TED2017
Joseph Redmon: How computers learn to recognize objects instantly
Joseph Redmon: Kako računalo uči trenutačno prepoznavati objekte
Filmed:
Readability: 4.5
2,471,805 views
Prije deset godina, istraživači su mislili da je gotovo nemoguće naučiti računalo prepoznati razliku između mačke i psa. Danas sustavi računalnog vida to rade s više od 99 posto točnosti. Kako? Joseph Redmon radi na sustavu YOLO (You Only Look Once), metodi detekcije objekata otvorenog koda koja može prepoznati predmete u slikama i videozapisima - od zebre do zaustavnih znakova - munjevitom brzinom. U izvanrednom demo prikazu, Redmon pokazuje ovaj važan korak naprijed za aplikacije poput autonomnih automobila, robotike, pa čak i otkrivanja raka.
Joseph Redmon - Computer scientist
Joseph Redmon works on the YOLO algorithm, which combines the simple face detection of your phone camera with a cloud-based AI -- in real time. Full bio
Joseph Redmon works on the YOLO algorithm, which combines the simple face detection of your phone camera with a cloud-based AI -- in real time. Full bio
Double-click the English transcript below to play the video.
00:12
Ten years ago,
0
825
1151
Prije deset godina,
istraživači računalnog vida mislili su
da je naučiti računalo
da je naučiti računalo
00:14
computer vision researchers
thought that getting a computer
thought that getting a computer
1
2000
2776
kako razlikovati između mačke i psa
00:16
to tell the difference
between a cat and a dog
between a cat and a dog
2
4800
2696
00:19
would be almost impossible,
3
7520
1976
gotovo nemoguće,
00:21
even with the significant advance
in the state of artificial intelligence.
in the state of artificial intelligence.
4
9520
3696
čak i uz značajan napredak
u razvoju umjetne inteligencije.
u razvoju umjetne inteligencije.
00:25
Now we can do it at a level
greater than 99 percent accuracy.
greater than 99 percent accuracy.
5
13240
3560
Sad to možemo učiniti
s više od 99 posto točnosti.
s više od 99 posto točnosti.
00:29
This is called image classification --
6
17680
1856
To se zove klasifikacija slike -
00:31
give it an image,
put a label to that image --
put a label to that image --
7
19560
3096
dati sliku, staviti oznaku na sliku -
00:34
and computers know
thousands of other categories as well.
thousands of other categories as well.
8
22680
3040
a računala znaju
i tisuće drugih kategorija.
i tisuće drugih kategorija.
00:38
I'm a graduate student
at the University of Washington,
at the University of Washington,
9
26680
2896
Ja sam postdiplomac
na Sveučilištu u Washingtonu
na Sveučilištu u Washingtonu
00:41
and I work on a project called Darknet,
10
29600
1896
i radim na projektu pod nazivom Darknet,
00:43
which is a neural network framework
11
31520
1696
što je neuronska mrežna struktura
00:45
for training and testing
computer vision models.
computer vision models.
12
33240
2816
za obuku i testiranje
modela računalnog vida.
modela računalnog vida.
00:48
So let's just see what Darknet thinks
13
36080
2976
Pa pogledajmo što Darknet misli
00:51
of this image that we have.
14
39080
1760
o ovoj slici koju imamo.
00:54
When we run our classifier
15
42520
2336
Kad smo pokrenuti naš klasifikator
00:56
on this image,
16
44880
1216
na ovoj slici,
00:58
we see we don't just get
a prediction of dog or cat,
a prediction of dog or cat,
17
46120
2456
ne dobivamo samo predviđanja
je li to pas ili mačka,
je li to pas ili mačka,
01:00
we actually get
specific breed predictions.
specific breed predictions.
18
48600
2336
nego čak i određena predviđanja pasmine.
01:02
That's the level
of granularity we have now.
of granularity we have now.
19
50960
2176
To je razina zrnatosti koju imamo sada.
01:05
And it's correct.
20
53160
1616
I to je točno.
01:06
My dog is in fact a malamute.
21
54800
1840
Moj pas je doista malamut.
01:09
So we've made amazing strides
in image classification,
in image classification,
22
57040
4336
Napravili smo nevjerojatne pomake
u klasifikaciji slike,
u klasifikaciji slike,
ali što se događa kad
pokrenemo klasifikator
pokrenemo klasifikator
01:13
but what happens
when we run our classifier
when we run our classifier
23
61400
2000
01:15
on an image that looks like this?
24
63424
1960
na sliku koja izgleda ovako?
01:19
Well ...
25
67080
1200
Dobro ...
01:24
We see that the classifier comes back
with a pretty similar prediction.
with a pretty similar prediction.
26
72640
3896
Vidimo da je klasifikator vraća
uz prilično slična predviđanja.
uz prilično slična predviđanja.
01:28
And it's correct,
there is a malamute in the image,
there is a malamute in the image,
27
76560
3096
I to je točno, na slici je malamut,
01:31
but just given this label,
we don't actually know that much
we don't actually know that much
28
79680
3696
ali samo s tom oznakom
ne znamo mnogo
ne znamo mnogo
01:35
about what's going on in the image.
29
83400
1667
o tome što se događa na slici.
01:37
We need something more powerful.
30
85091
1560
Trebamo nešto snažnije.
01:39
I work on a problem
called object detection,
called object detection,
31
87240
2616
Radim na problemu koji se zove
otkrivanje objekta,
otkrivanje objekta,
01:41
where we look at an image
and try to find all of the objects,
and try to find all of the objects,
32
89880
2936
gdje gledamo sliku i
pokušavamo pronaći sve objekte,
pokušavamo pronaći sve objekte,
01:44
put bounding boxes around them
33
92840
1456
staviti okvire oko njih
01:46
and say what those objects are.
34
94320
1520
i reći ono što ti predmeti su.
01:48
So here's what happens
when we run a detector on this image.
when we run a detector on this image.
35
96400
3280
Evo što se događa kad pokrenemo
detektor na ovoj slici.
detektor na ovoj slici.
01:53
Now, with this kind of result,
36
101240
2256
Ovakvom vrstom rezultata
možemo napraviti puno više
s algoritmima računalnog vida.
s algoritmima računalnog vida.
01:55
we can do a lot more
with our computer vision algorithms.
with our computer vision algorithms.
37
103520
2696
01:58
We see that it knows
that there's a cat and a dog.
that there's a cat and a dog.
38
106240
2976
Vidimo da zna da su tu mačka i pas.
02:01
It knows their relative locations,
39
109240
2256
Zna njihove relativne položaje,
02:03
their size.
40
111520
1216
njihovu veličinu.
02:04
It may even know some extra information.
41
112760
1936
Čak može znati neke dodatne informacije.
U pozadini je knjiga.
02:06
There's a book sitting in the background.
42
114720
1960
02:09
And if you want to build a system
on top of computer vision,
on top of computer vision,
43
117280
3256
Ako želite izgraditi sustav
na osnovi računalnog vida,
na osnovi računalnog vida,
02:12
say a self-driving vehicle
or a robotic system,
or a robotic system,
44
120560
3456
recimo autonomno vozilo
ili robotski sustav,
ili robotski sustav,
02:16
this is the kind
of information that you want.
of information that you want.
45
124040
2456
ovo je vrsta informacija koje želite.
02:18
You want something so that
you can interact with the physical world.
you can interact with the physical world.
46
126520
3239
Želite nešto da možete
komunicirati s fizičkim svijetom.
komunicirati s fizičkim svijetom.
02:22
Now, when I started working
on object detection,
on object detection,
47
130759
2257
Kad sam počeo raditi na
prepoznavanju objekata,
prepoznavanju objekata,
02:25
it took 20 seconds
to process a single image.
to process a single image.
48
133040
3296
trebalo je 20 sekundi
za obradu jedne slike.
za obradu jedne slike.
02:28
And to get a feel for why
speed is so important in this domain,
speed is so important in this domain,
49
136360
3880
A kako biste dobili osjećaj zašto je
brzina ovdje tako važna,
brzina ovdje tako važna,
02:33
here's an example of an object detector
50
141120
2536
evo primjera detektora objekta
02:35
that takes two seconds
to process an image.
to process an image.
51
143680
2416
koji treba dvije sekunde za obradu slike.
02:38
So this is 10 times faster
52
146120
2616
Dakle ovo je 10 puta brže
02:40
than the 20-seconds-per-image detector,
53
148760
3536
od detektora kojem treba
20 sekundi po slici,
20 sekundi po slici,
i možete vidjeti da se za vrijeme
dok on učini predviđanja,
dok on učini predviđanja,
02:44
and you can see that by the time
it makes predictions,
it makes predictions,
54
152320
2656
02:47
the entire state of the world has changed,
55
155000
2040
promijenilo čitavo stanje u svijetu,
02:49
and this wouldn't be very useful
56
157880
2416
i to ne bi bilo vrlo korisno
02:52
for an application.
57
160320
1416
za neku primjenu.
02:53
If we speed this up
by another factor of 10,
by another factor of 10,
58
161760
2496
Ako ovo gore ubrzamo
još jednom za faktor 10,
još jednom za faktor 10,
02:56
this is a detector running
at five frames per second.
at five frames per second.
59
164280
2816
to je detektor koji radi
na pet sličica u sekundi.
na pet sličica u sekundi.
02:59
This is a lot better,
60
167120
1536
To je puno bolje,
03:00
but for example,
61
168680
1976
ali, na primjer,
03:02
if there's any significant movement,
62
170680
2296
ako postoji bilo kakav značajan pokret,
03:05
I wouldn't want a system
like this driving my car.
like this driving my car.
63
173000
2560
ne bih želio da sustav poput ovog
vozi moj auto.
vozi moj auto.
03:09
This is our detection system
running in real time on my laptop.
running in real time on my laptop.
64
177120
3240
Ovo je naš sustav otkrivanja
u realnom vremenu na mom laptopu.
u realnom vremenu na mom laptopu.
03:13
So it smoothly tracks me
as I move around the frame,
as I move around the frame,
65
181000
3136
Glatko me prati kako se krećem kroz kadar,
03:16
and it's robust to a wide variety
of changes in size,
of changes in size,
66
184160
3720
i otporan je na razne promjene veličine,
03:21
pose,
67
189440
1200
položaja,
03:23
forward, backward.
68
191280
1856
naprijed, natrag.
03:25
This is great.
69
193160
1216
Ovo je super.
03:26
This is what we really need
70
194400
1736
To je ono što stvarno trebamo
03:28
if we're going to build systems
on top of computer vision.
on top of computer vision.
71
196160
2896
ako ćemo graditi sustave
na osnovi računalnog vida,
na osnovi računalnog vida,
03:31
(Applause)
72
199080
4000
(Pljesak)
03:36
So in just a few years,
73
204280
2176
U samo nekoliko godina
03:38
we've gone from 20 seconds per image
74
206480
2656
došli smo od 20 sekundi po slici
03:41
to 20 milliseconds per image,
a thousand times faster.
a thousand times faster.
75
209160
3536
do 20 milisekundi po slici,
tisuću puta brže.
tisuću puta brže.
03:44
How did we get there?
76
212720
1416
Kako smo došli dovde?
03:46
Well, in the past,
object detection systems
object detection systems
77
214160
3016
Nekada su sustavi za otkrivanje predmeta
03:49
would take an image like this
78
217200
1936
uzimali sliku poput ove
03:51
and split it into a bunch of regions
79
219160
2456
i podijelili je na hrpu područja
03:53
and then run a classifier
on each of these regions,
on each of these regions,
80
221640
3256
i zatim pokrenuli klasifikator
na svakom od tih područja.
na svakom od tih područja.
03:56
and high scores for that classifier
81
224920
2536
Visoki rezultati za taj klasifikator
03:59
would be considered
detections in the image.
detections in the image.
82
227480
3136
smatrali su se detekcijom u slici.
04:02
But this involved running a classifier
thousands of times over an image,
thousands of times over an image,
83
230640
4056
No, to je značilo rad klasifikatora
tisuće puta na slici,
tisuće puta na slici,
04:06
thousands of neural network evaluations
to produce detection.
to produce detection.
84
234720
2920
tisuće procjena neuronskih mreža
kako bi dobili detekciju.
kako bi dobili detekciju.
04:11
Instead, we trained a single network
to do all of detection for us.
to do all of detection for us.
85
239240
4536
Umjesto toga smo naučili jednu mrežu
da učini sve detekcije za nas.
da učini sve detekcije za nas.
04:15
It produces all of the bounding boxes
and class probabilities simultaneously.
and class probabilities simultaneously.
86
243800
4280
Ona istodobno proizvodi sve okvire
i klase vjerojatnosti.
i klase vjerojatnosti.
04:20
With our system, instead of looking
at an image thousands of times
at an image thousands of times
87
248680
3496
S našim sustavom, umjesto da
gledate sliku tisuće puta
gledate sliku tisuće puta
04:24
to produce detection,
88
252200
1456
kako bi postigao detekciju,
04:25
you only look once,
89
253680
1256
gledate samo jednom,
04:26
and that's why we call it
the YOLO method of object detection.
the YOLO method of object detection.
90
254960
2920
zato ga zovemo YOLO metoda
za detekciju objekta.
za detekciju objekta.
04:31
So with this speed,
we're not just limited to images;
we're not just limited to images;
91
259360
3976
Dakle, ovom brzinom
nismo ograničeni samo na slike;
nismo ograničeni samo na slike;
04:35
we can process video in real time.
92
263360
2416
možemo obraditi video u realnom vremenu.
04:37
And now, instead of just seeing
that cat and dog,
that cat and dog,
93
265800
3096
Sad, umjesto da samo vidimo mačku i psa,
04:40
we can see them move around
and interact with each other.
and interact with each other.
94
268920
2960
vidimo kako se kreću
i međusobno komuniciraju.
i međusobno komuniciraju.
04:46
This is a detector that we trained
95
274560
2056
To je detektor koji smo obučili
04:48
on 80 different classes
96
276640
4376
na 80 različitih klasa
04:53
in Microsoft's COCO dataset.
97
281040
3256
u Microsoftovoj zbirci podataka COCO.
04:56
It has all sorts of things
like spoon and fork, bowl,
like spoon and fork, bowl,
98
284320
3336
Ona ima svašta, poput žlice
i vilice, zdjele,
i vilice, zdjele,
04:59
common objects like that.
99
287680
1800
obične predmete poput tih.
05:02
It has a variety of more exotic things:
100
290360
3096
Ima raznih egzotičnijih stvari:
05:05
animals, cars, zebras, giraffes.
101
293480
3256
životinje, automobili, zebre, žirafe.
05:08
And now we're going to do something fun.
102
296760
1936
A sada idemo učiniti nešto zabavno.
05:10
We're just going to go
out into the audience
out into the audience
103
298720
2096
Samo ćemo otići u publiku
05:12
and see what kind of things we can detect.
104
300840
2016
i vidjeti što možemo otkriti.
05:14
Does anyone want a stuffed animal?
105
302880
1620
Želi li tko plišanu životinju?
05:18
There are some teddy bears out there.
106
306000
1762
Tamo ima nekih medvjedića.
05:22
And we can turn down
our threshold for detection a little bit,
our threshold for detection a little bit,
107
310040
4536
Možemo malo smanjiti prag detekcije,
05:26
so we can find more of you guys
out in the audience.
out in the audience.
108
314600
3400
tako da možemo naći više vas u publici.
05:31
Let's see if we can get these stop signs.
109
319560
2336
Da vidimo možemo li dobiti
ove znakove STOP.
ove znakove STOP.
05:33
We find some backpacks.
110
321920
1880
Nalazimo neke ruksake.
05:37
Let's just zoom in a little bit.
111
325880
1840
Zumirajmo samo malo.
05:42
And this is great.
112
330320
1256
I to je super.
05:43
And all of the processing
is happening in real time
is happening in real time
113
331600
3176
Sva obrada se događa u stvarnom vremenu
05:46
on the laptop.
114
334800
1200
na laptopu.
05:49
And it's important to remember
115
337080
1456
I to je važno zapamtiti
da je ovo sustav za detekciju objekta
opće namjene,
opće namjene,
05:50
that this is a general purpose
object detection system,
object detection system,
116
338560
3216
05:53
so we can train this for any image domain.
117
341800
5000
možemo ga trenirati za bilo koju domenu.
06:00
The same code that we use
118
348320
2536
Isti kod koji koristimo
06:02
to find stop signs or pedestrians,
119
350880
2456
za pronaći znakove STOP ili pješake,
06:05
bicycles in a self-driving vehicle,
120
353360
1976
bicikle u autonomnim vozilima,
06:07
can be used to find cancer cells
121
355360
2856
može se koristiti
kako bi pronašli stanice raka
kako bi pronašli stanice raka
06:10
in a tissue biopsy.
122
358240
3016
u biopsiji tkiva.
06:13
And there are researchers around the globe
already using this technology
already using this technology
123
361280
4040
A znanstvenici diljem svijeta već
koriste ovu tehnologiju
koriste ovu tehnologiju
06:18
for advances in things
like medicine, robotics.
like medicine, robotics.
124
366240
3416
za napredak u medicini, robotici.
06:21
This morning, I read a paper
125
369680
1376
Jutros sam pročitao članak
06:23
where they were taking a census
of animals in Nairobi National Park
of animals in Nairobi National Park
126
371080
4576
o popisu životinja u
Nacionalnom parku Nairobi
Nacionalnom parku Nairobi
06:27
with YOLO as part
of this detection system.
of this detection system.
127
375680
3136
koristeći YOLO u sustavu detekcije.
06:30
And that's because Darknet is open source
128
378840
3096
A to je zato što je Darknet open source,
06:33
and in the public domain,
free for anyone to use.
free for anyone to use.
129
381960
2520
u javnoj domeni, besplatan
svakomu za korištenje.
svakomu za korištenje.
06:37
(Applause)
130
385600
5696
(Pljesak)
06:43
But we wanted to make detection
even more accessible and usable,
even more accessible and usable,
131
391320
4936
No, željeli smo napraviti detekciju
još dostupnijom i korisnijom
još dostupnijom i korisnijom
06:48
so through a combination
of model optimization,
of model optimization,
132
396280
4056
pa smo kombinacijom optimizacije modela,
06:52
network binarization and approximation,
133
400360
2296
binarizacije mreže i aproksimacije
06:54
we actually have object detection
running on a phone.
running on a phone.
134
402680
3920
dobili detekciju objekata
koja radi na mobitelu.
koja radi na mobitelu.
07:04
(Applause)
135
412800
5320
(Pljesak)
07:10
And I'm really excited because
now we have a pretty powerful solution
now we have a pretty powerful solution
136
418960
5056
A ja sam stvarno uzbuđen
jer sada imamo moćno rješenje
jer sada imamo moćno rješenje
07:16
to this low-level computer vision problem,
137
424040
2296
problema računalnog vida
na osnovnoj razini,
na osnovnoj razini,
07:18
and anyone can take it
and build something with it.
and build something with it.
138
426360
3856
i svatko ga može uzeti
i graditi nešto njime.
i graditi nešto njime.
07:22
So now the rest is up to all of you
139
430240
3176
Sad je sve ostalo do vas
07:25
and people around the world
with access to this software,
with access to this software,
140
433440
2936
i ljudi diljem svijeta
s pristupom tom softveru,
s pristupom tom softveru,
07:28
and I can't wait to see what people
will build with this technology.
will build with this technology.
141
436400
3656
jedva čekam vidjeti što će
ljudi učiniti s ovom tehnologijom.
ljudi učiniti s ovom tehnologijom.
07:32
Thank you.
142
440080
1216
Hvala vam.
(Pljesak)
07:33
(Applause)
143
441320
3440
ABOUT THE SPEAKER
Joseph Redmon - Computer scientistJoseph Redmon works on the YOLO algorithm, which combines the simple face detection of your phone camera with a cloud-based AI -- in real time.
Why you should listen
Computer scientist Joseph Redmon is working on the YOLO (You Only Look Once) algorithm, which has a simple goal: to deliver image recognition and object detection at a speed that would seem science-fictional only a few years ago. The algorithm looks like the simple face detection of a camera app but with the level complexity of systems like Google's Deep Mind Cloud Vision, using Convolutional Deep Neural Networks to crunch object detection in realtime. It's the kind of technology that will be embedded on all smartphones in the next few years.
Redmon is also internet-famous for his resume.
Joseph Redmon | Speaker | TED.com