<s>
Computer	B-Application
vision	I-Application
research	O
is	O
driven	O
by	O
standard	O
evaluation	O
practices	O
.	O
</s>
<s>
The	O
current	O
systems	O
are	O
tested	O
by	O
their	O
accuracy	O
for	O
tasks	O
like	O
object	B-General_Concept
detection	I-General_Concept
,	O
segmentation	B-Algorithm
and	O
localization	O
.	O
</s>
<s>
Methods	O
like	O
the	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
seem	O
to	O
be	O
doing	O
pretty	O
well	O
in	O
these	O
tasks	O
,	O
but	O
the	O
current	O
systems	O
are	O
still	O
not	O
close	O
to	O
solving	O
the	O
ultimate	O
problem	O
of	O
understanding	O
images	O
the	O
way	O
humans	O
do	O
.	O
</s>
<s>
have	O
introduced	O
the	O
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
for	O
computer	B-Application
vision	I-Application
systems	O
.	O
</s>
<s>
The	O
test	O
is	O
only	O
about	O
vision	O
and	O
does	O
not	O
require	O
any	O
natural	B-Language
language	I-Language
processing	I-Language
.	O
</s>
<s>
Research	O
in	O
computer	B-Application
vision	I-Application
dates	O
back	O
to	O
the	O
1960s	O
when	O
Seymour	O
Papert	O
first	O
attempted	O
to	O
solve	O
the	O
problem	O
.	O
</s>
<s>
The	O
reason	O
why	O
it	O
was	O
not	O
successful	O
was	O
because	O
computer	B-Application
vision	I-Application
is	O
more	O
complicated	O
than	O
what	O
people	O
think	O
.	O
</s>
<s>
Perceptrons	B-Algorithm
by	O
Frank	O
Rosenblatt	O
,	O
which	O
is	O
a	O
form	O
of	O
the	O
neural	B-Architecture
networks	I-Architecture
,	O
was	O
one	O
of	O
the	O
first	O
such	O
approaches	O
.	O
</s>
<s>
These	O
simple	O
neural	B-Architecture
networks	I-Architecture
could	O
not	O
live	O
up	O
to	O
their	O
expectations	O
and	O
had	O
certain	O
limitations	O
due	O
to	O
which	O
they	O
were	O
not	O
considered	O
in	O
future	O
research	O
.	O
</s>
<s>
Later	O
with	O
the	O
availability	O
of	O
the	O
hardware	O
and	O
some	O
processing	O
power	O
the	O
research	O
shifted	O
to	O
image	B-Algorithm
processing	I-Algorithm
which	O
involves	O
pixel-level	O
operations	O
,	O
like	O
finding	B-Algorithm
edges	I-Algorithm
,	O
de-noising	O
images	O
or	O
applying	O
filters	O
to	O
name	O
a	O
few	O
.	O
</s>
<s>
During	O
this	O
time	O
the	O
neural	B-Architecture
networks	I-Architecture
also	O
resurfaced	O
as	O
it	O
was	O
shown	O
that	O
the	O
limitations	O
of	O
the	O
perceptrons	B-Algorithm
can	O
be	O
overcome	O
by	O
Multi-layer	B-Algorithm
perceptrons	I-Algorithm
.	O
</s>
<s>
Also	O
in	O
the	O
early	O
1990s	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
were	O
born	O
which	O
showed	O
great	O
results	O
on	O
digit	O
recognition	O
but	O
did	O
not	O
scale	O
up	O
well	O
on	O
harder	O
problems	O
.	O
</s>
<s>
The	O
late	O
1990s	O
and	O
early	O
2000s	O
saw	O
the	O
birth	O
of	O
modern	O
computer	B-Application
vision	I-Application
.	O
</s>
<s>
One	O
of	O
the	O
reasons	O
this	O
happened	O
was	O
due	O
to	O
the	O
availability	O
of	O
key	O
,	O
feature	B-Algorithm
extraction	I-Algorithm
and	O
representation	O
algorithms	O
.	O
</s>
<s>
This	O
led	O
to	O
the	O
emergence	O
of	O
challenges	O
like	O
the	O
Pascal	O
VOC	O
challenge	O
and	O
the	O
ImageNet	B-General_Concept
challenge	O
.	O
</s>
<s>
Better	O
algorithms	O
were	O
introduced	O
for	O
specific	O
tasks	O
like	O
object	B-General_Concept
detection	I-General_Concept
and	O
classification	O
.	O
</s>
<s>
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
aims	O
to	O
give	O
a	O
new	O
direction	O
to	O
the	O
computer	B-Application
vision	I-Application
research	O
which	O
would	O
lead	O
to	O
the	O
introduction	O
of	O
systems	O
that	O
will	O
be	O
one	O
step	O
closer	O
to	O
understanding	O
images	O
the	O
way	O
humans	O
do	O
.	O
</s>
<s>
One	O
of	O
the	O
most	O
famous	O
datasets	O
in	O
computer	B-Application
vision	I-Application
is	O
which	O
is	O
used	O
to	O
assess	O
the	O
problem	O
of	O
object	O
level	O
Image	O
classification	O
.	O
</s>
<s>
ImageNet	B-General_Concept
is	O
one	O
of	O
the	O
largest	O
annotated	O
datasets	O
available	O
and	O
has	O
over	O
one	O
million	O
images	O
.	O
</s>
<s>
The	O
other	O
important	O
vision	O
task	O
is	O
object	B-General_Concept
detection	I-General_Concept
and	O
localisation	O
which	O
refers	O
to	O
detecting	O
the	O
object	O
instance	O
in	O
the	O
image	O
and	O
providing	O
the	O
bounding	O
box	O
coordinates	O
around	O
the	O
object	O
instance	O
or	O
segmenting	O
the	O
object	O
.	O
</s>
<s>
The	O
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
(	O
VTT	O
)	O
unlike	O
the	O
Turing	O
test	O
has	O
a	O
query	O
engine	O
system	O
which	O
interrogates	O
a	O
computer	B-Application
vision	I-Application
system	O
in	O
the	O
presence	O
of	O
a	O
human	O
co-ordinator	O
.	O
</s>
<s>
The	O
aim	O
of	O
the	O
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
is	O
to	O
evaluate	O
the	O
Image	B-Application
understanding	I-Application
of	O
a	O
computer	O
system	O
,	O
and	O
an	O
important	O
part	O
of	O
image	B-Application
understanding	I-Application
is	O
the	O
story	O
line	O
of	O
the	O
image	O
.	O
</s>
<s>
Additionally	O
all	O
of	O
this	O
vocabulary	O
is	O
used	O
in	O
context	O
of	O
rectangular	O
image	B-Algorithm
regions	I-Algorithm
w	O
\in	O
W	O
which	O
allow	O
for	O
the	O
localisation	O
of	O
objects	O
in	O
the	O
image	O
.	O
</s>
<s>
As	O
mentioned	O
before	O
the	O
core	O
of	O
the	O
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
is	O
the	O
query	O
generator	O
which	O
generates	O
a	O
sequence	O
of	O
binary	O
questions	O
such	O
that	O
the	O
answer	O
to	O
any	O
question	O
k	O
is	O
unpredictable	O
given	O
the	O
correct	O
answers	O
to	O
the	O
previous	O
k	O
−	O
1	O
questions	O
.	O
</s>
<s>
But	O
they	O
propose	O
a	O
different	O
version	O
of	O
the	O
visual	B-General_Concept
Turing	I-General_Concept
test	I-General_Concept
which	O
takes	O
on	O
a	O
holistic	O
approach	O
and	O
expects	O
the	O
participating	O
system	O
to	O
exhibit	O
human	O
like	O
common	O
sense	O
.	O
</s>
<s>
It	O
evaluates	O
how	O
the	O
computer	B-Application
vision	I-Application
systems	O
understand	O
the	O
Images	O
as	O
compared	O
to	O
humans	O
.	O
</s>
<s>
The	O
Visual	B-General_Concept
Turing	I-General_Concept
Test	I-General_Concept
is	O
expected	O
to	O
give	O
a	O
new	O
direction	O
to	O
the	O
computer	B-Application
vision	I-Application
research	O
.	O
</s>
<s>
Companies	O
like	O
Google	B-Application
and	O
Facebook	B-Application
are	O
investing	O
millions	O
of	O
dollars	O
into	O
computer	B-Application
vision	I-Application
research	O
,	O
and	O
are	O
trying	O
to	O
build	O
systems	O
that	O
closely	O
resemble	O
the	O
human	O
visual	O
system	O
.	O
</s>
<s>
Recently	O
Facebook	B-Application
announced	O
its	O
new	O
platform	O
M	O
,	O
which	O
looks	O
at	O
an	O
image	O
and	O
provides	O
a	O
description	O
of	O
it	O
to	O
help	O
the	O
visually	O
impaired	O
.	O
</s>
