Practical 1.2 – image package Video Lecture - Torch A Practical Hands

we will have now another view on ports
image package image is the torch 7
package for processing images it allows
us to save and load JPEG PNG pbn and
pgms
allow us to do some easy transformations
like translation scaling rotation or its
on the vertical flip we can also do
parameterize transformations like
convolutions and warping you can also
draw some text and rectangles on images
and we are provided with a graphical
user interface for displaying our images
moreover we can perform color space
conversions from into our GPS Yub lab
and HSL finally it provides us with
constructors for creating Malena
gaussian and laplacian kernels we will
see the following functions we will
start with lower aux list and install to
check whether the env package is
installed in our system then we will
meet our new wrapper which allow us to
use the Qt libraries and we will use it
with the nth library then we will learn
how to load an image we will learn also
the equal operator in the Q Lua
interpreter we will learn how to scale
an image how to display an image how to
make a copy a deep copy of an image how
to replace the content of a plane for
example with a specific value moreover
we will see a more advanced display
where we can specify a title for the
window zoom factor the number of
multiple images we would like to show
the arrow and how much space you would
like to put between those images
moreover we see how to extract this
collection of images we just displayed
we see how to save an image in more
detail a PNG file how to crop an image
we will see the names of some popular
color space conversion
and then finally we see how to draw text
on an image and how to draw a rectangle
more precisely it's going to be a
bounding box around Lina's face let's
say we have an image we would like to
resize in this tutorial instead of using
the red 4 th we have been using so far
we will be using q1 which is a cutie
enable blueish shell we should check
that the env package is installed on our
machine we can do so by typing the wire
rocks list e NV and we can see that the
env
is installed on our machine so let's
open q1 - ad e NV to load the library
env which provides our report with nice
functionalities we can start by
requiring the package image then we can
have a variable AI equal image dot load
and then the name of the file AI dot jpg
in kurwa we have to type an equal in
order to have a result print on screen
so let's print the size of a I AI has
three planes RGB 1000 and 1500 columns
the type of AI it's a double tensor
let's say we would like to nicely
visualize this image on our screen let's
query the current resolution of my
screen on OS X I can do so by typing
system profile SP display
datatype and I can grab the resolution
and we have that the resolution is of
900 rolls and 1440 columns so we had it
our image it's higher and wider than the
car in the screen let's resize our image
to half of its dimension so we can type
AI dot scale equal image dot scale of
our image AI we can say half of it
throws and half of its height if we
check now the dimensionality of a scale
it's going to be three planes 500 rolls
and 750 columns we can now easily
display this image on the screen I can
do so by typing image display and ai
scale and here is our image let's now
check the content of green in this image
to do so let's go back to the terminal
let's have AI green equal a I scale
clone in this way we made a deep clone
of the AI scale tensor otherwise you
would have sure simply the same
reference to the same tensor if you
would have simply used the equal
assignment operator so AI green again
it's three planes 500 750 columns let's
take the first plane
which is the red plane and fill it with
zeros then RGB so let's take the third
plane and also fill it to zeros so now
AI green will have only content relative
to the green channel let's visualize now
the content of AI green so image dot
display and I could simply type AI green
this time we change and we write curly
bracket and then we have image equal Adi
green legend equal green plane let's
check the result here we have the green
content of the previous image let's say
we would like to visualize both image on
the same window we can do so by typing
image display image equal the table of
AI scale and AI green then we have
legend equal original and green plane
and then let's put some equal or 0.5
otherwise le won't fit let's check the
result and here they are the original on
the left hand side and the green plane
on the right hand side let's display as
well the other two channels perhaps
below and on the right hand side let's
go back to the terminal
so let's have a I read equal toward
zeros of the size of a I scale and then
we have that we copy over just the first
plane equal a I scale first claim and
then we can have a I blue equal torch
dot zeroes a I scale and then a I blow
our GP so B is the last one equal to AI
scale number three and now we can
display all the four images so we can do
image dot display cut the bracket and
let's have image equal a I scale and now
we have a I RGB so our then we have AI g
green and now we have a I blow and then
let's put again the zone equal zero
point five and let's set the number of
rows to be two at most and then we write
the legend which is going to be original
and then we have
read every below we have G and P close
and let's check the result and here we
have the four maps so we have the
original image in the top left corner
the red plain top right we can see that
in the top right area there is little
red where we have lots of green in this
area and as well here that we have
mainly blue we have less red here the
same way where we have more red here we
can see that there is less screen and
then we can see in the yellow area there
is basically and in the red area there
are no whatsoever blue values you can
still see some spots here and here
because they correspond to the bright
white of those spots let's say we are
satisfied with our result and we would
like to save the content of this screen
so let's go back we can convert this
collection of images into one image by
using the a very similar function let's
go to the beginning and instead of
having display we have image to display
tensor instead of having image we put
input and then at the beginning we are
going to have our output image and we
can press ENTER now if we would like to
display to be sure it is correct we can
do image dot display image equal output
image and so equals 0.5 and Legend equal
Oh let's check so here we have the D
output image are those four brains glued
together let's save this image on our
desktop so let's go back to the terminal
we can do so by issuing the command
image dot save let's save in PNG perhaps
initially we open a jpg image and now we
are saving a PNG so we can easily
convert among different images format so
let's put here the name and we have 4
glued brains dot PNG and then we can
insert our image we won't like to say we
said we would like to save it on the
desktop so we can do this top slash for
glued brains and if we go back on the
desktop we can see here that we have our
4 glue raines-ing and we can open it
let's say now we would like to extract
the top-left quadrant of our newly
generated image so let's print first the
size of our image which is three planes
an RGB image so a color image with 1000
lines and 1500 columns itself now our
crop image equal image dot crop of our
output image from the 0 0 coordinate so
top left quadrant to the 1500 divided by
2 x coordinate and 1000 divided by 2 y
coordinate we can check now the
dimensionality which is exactly a color
image so three planes 500 rows and 750
columns we can now display this image
to be sure we have cropped the right one
so image display image equal crop image
and then legend equal crop and we can
check the result now and there we go the
cropped image there are many more
transformation we can apply to our image
we can have a list of all
transformations by typing image dot tap
top we can see crop we can see Alexander
flip we can see vertical flip can also
see HSL to RGB and HSV to RGB or RGB to
HSN RGB to HSV and RGB to lab also RGB
to n our GP and RGB to wide so black and
white and RGB to Yub we can rotate the
email we can save in different format
like JPEG PGM P&G and ppm as we saw we
can scale our image and then also one
interesting one is from a grayscale to a
jet color map I wrote some time ago
let's finish now by trying the Lena
function which returns an image that has
been widely used in the computer vision
field we can simply display this image
by typing image dot display and then we
have image dot Lena and here she is
finally let's see how to write some text
and draw some rectangles on our images
this is very important for writing
labels and bounding boxes image draw
text on image Lena
and we have hello new line word add the
coordinate 1010 and here we have image
dot this play and if we check the result
we have Lena and string hello world in
the top-left corner at the position 10
pick certain pixel we can change now
for example the position and color and
size so let's do the same and in here
you can do 20 for example and then we
have older equal 0 255 0 and then we
have size equal 5 close and if we check
the result we have here our hello world
in a green and bigger character we can
try one more variation so let's go up
and then let's also add here the ground
color equal HAP's
red to 5500 let's press Enter
and here we have the third variation
green font on red the ground terrible
finally we can learn how to draw a
bounding box so let's go back here image
dot display and open parenthesis then
let's go in
I draw rectangle and then we have emails
dot Dana coordinate 200 200
and then 317 and 400 and we have some
specifications for example can help line
with equal 5 and now we can have color
equal perhaps screen and so let's close
this guy and that goes listen and let's
close this and let's see our result so
there we go we have a bounding box on
Lina's face