►
From YouTube: DevoWorm Summer of Code weekly meeting, 7-19
Description
Meeting for Week 11. Asmit Singh, Ujjwal Singh, Paras Mehan present Machine Learning results for the Digital Bacillaria.
A
You
know
a
health
problem
that
he
informed
you
about
by
email
and
he's
been
recovering,
he's
back
in
I,
guess
in
good
condition.
He's
on
a
treadmill
today
he's
joining
us
from
is
working
treadmill,
so
we
and
I'm
the
deso
area
stuff
and
so
welcome
back
okay.
So
anyone
does
anyone
have
any
questions
that
they
need
to
talk
about
before
we
get
started.
A
B
A
B
B
B
C
A
A
B
A
A
B
B
A
A
D
C
B
A
Mean
it's:
it's
probably
a
matter
of
like
I.
Don't
know
if
it's
like
you
know
finding
something
from
like
the
black
most
cell
to
the
right
most
selfie
like
use
the
cell
centroids.
You
might
be
able
to
do
it,
but
first
thing
overlap.
So
it's
a
bit
tricky,
but
then
you
know
there
might
be
a
easy
way
to
do
it.
Just
by
some.
A
You
know
measure
that
you're
already
outputting
and
just
say
like
if
it's
you
know
take
all
these
that
are
in
the
same
frame
in
the
same
image
and
order
them
in
some
way
and
then
output
like
an
order,
and
then
you
would
have
like
an
address,
so
it
would
be
like
each
image
would
have
its
own
order,
and
so
they
go
from
image
to
image.
You
would
say
well.
This
was
number
one
an
image
a
this
is
now
number
one
an
image
B
and
then
you
can
actually
track.
E
A
I
just
had
a
question
about
like
we
were
talking
and
it's
like
about
different
algorithms
you're
using
two
segments
and
for
a
while,
you
were
getting
like
a
result
where
you
were
getting
edges,
really
fuzzy.
You
couldn't
go
in
lock
onto
the
edges,
and
these
images
look
a
lot
better
in
terms
of
like
the
boundaries
they're
a
lot
sharper
and
they
seem
to
pick
up
a
lot
more
of
the
detail,
at
least
in
terms
of
the
filament
boundary
and
like
distinguishing
between
background
and
the
actual
boundary.
So
what?
What
was
it
that
you
found?
A
E
C
C
E
C
A
Okay,
there's
another
question
from
Richard:
please
plot
number
of
cells
versus
frame
number
just
identify
portions
of
a
movie
that
could
be
reliably
analyzed,
so
that
would
just
be
like
once
you
get
the
number
of
once
you
get
the
identity
of
the
cells.
So
we'd
have
like
this,
the
frame
number
and
the
cell
number
in
the
frame
ordered.
Then
we
could
take.
You
know
each.
We
could
just
basically
take
each
frame
and
see
how
many
cells
we
have
in
a
frame.
A
So
it's
a
count
of
how
many
number
of
cells
are
in
that
frame
and
then
we
could
plot
it
out
to
see
like
what
the
I
guess.
It's.
He
wants
to
see
what
cut
what
parts
of
the
movie
do.
We
have
enough
cells
in
so
that
we
can
track
them
across
frames
because
I
mean
you
know
they
move
around
in
and
out
of
the
frame.
So
I
don't
know
like
that.
That's
one
way
you
could
do
that
is
just
simply
produce
a
plot
by
gorilla.
E
A
E
A
So
I
mean
yeah,
so
part
of
that
and
I
mean
I,
don't
know
cuz,
they
do
move
in
and
out
of
the
frame.
So
it's
you
know
the
the
numbering
should
be
consistent
if
they
move
back
and
forth,
but
if
they
start
twisting
around
there
may
be
a
problem,
but
I
think
you
know
the
first
step.
There
is
just
to
say
like
to
go
through
the
video
and
see
if
the
results
you
know,
maybe
we
contract
one
cell
and
see
if
it's
you
know
follows
a
reasonable
path.
A
If
we
go
back
to
the
movie
and
look
at
it
and
then
look
you
know
we
can
plot
the
points
for
each.
You
know
like
cell
number,
one
in
each
frame.
Does
it
follow
what
the
movie
looks
like
and
then
we
can
plot
that
out.
We'd
also
do
like
an
aggregate
count.
So
I
mean
that's
those
are
ways
we
can
validate
the
actual
output.
A
I
know
I,
guess
I,
don't
know
what
you
mean
but
like
so
we'd
have
like,
because
you
have
where
we're
outputting
parameters
for
each
cell
in
each
frame.
But
then
you'd
have
like
over
a
large
number
of
frames.
You'd
have
all
these
data.
That
would
say
like
it's
cell,
one
in
image,
1
so
1
and
image,
2
and
so
forth,
and
what
we
want
to
do
is
we
want
to
be
able
to
validate
the
actuals.
A
So
we
want
to
be
able
to
take
those
output
parameters
and
actually
plot
them
in
like
a
bivariate
graph
and
then
look
at
the
trend
and
then
look
at
the
movies
like
just
go
back
and
validate
and
make
sure
that
it's
tracking
everything
now
in
terms
of
like,
if
you
get
to
a
point
where
some
of
the
cells
are
kind
of
you
know,
they're,
you
know
oblique
because
you
know
you're
taking
microscopy
images,
but
the
specimen
moves
with
respect
to
the
plane
of
view.
So
it's
not
perfect.
A
You
know
it's
not
like
you're
watching
something
move
on
a
two-dimensional
sheet
of
paper.
It's
you
know
on
a
flat
way.
It
can
twist
and
Bend
and
it's
you
know
they're
they're
ways
of
which
it
moves
that
may
not
be
able
to
be
detected
by
the
algorithm.
So
that's
you
know,
that's
why
we
want
to
do
a
lot
of
this
validation
stuff
but
in
terms
of
I
guess
in
terms
of
like
the
basic
parameters
like
size
and
that
sort
of
stuff
that
would
be
on
every
cell.
A
A
Yeah
I
mean
it's
in
some
of
the
movies
that
we
have
are
very
good
in
terms
of
like
you
know,
if
we
plug
it
into
something
that
we
can
train
the
model
on,
but
maybe
the
way
it's
you
know
the
the
quality
of
the
image,
because
you
have
a
lot
of
algae
moving
around
in
front
of
the
filaments
or
the
filaments
move,
I
guess
what
Allah
was
saying
was
that
the
filaments
move
and
then
you
know
as
they
move.
You
might
not
be
able
to
get
the
boundaries
quite
right,
yeah.
A
So
well,
that's
we'll
have
to
work
on
that.
A
little
bit
more
I
think
but
I
think
the
start.
We
should
just
get
the
parameters
and
maybe
make
some
plots
some
summary
plots
and
just
look
at
those
trends
and
look
at
the
movies
and
say:
does
this
make
sense?
You
know
the
data
makes
sense,
at
least
in
terms
of
like
reconstructing
the
devotion
or
the
position.
A
It's
not
going
to
be
perfect
and
there
might
be
some
false
positives
in
there,
but
I
think
and
I.
Don't
know
what
like
what
it
would
be
worth.
You
know
we
say
what
parts
of
what
parts
of
the
movie
be
worth.
Analyzing
I
mean
you
know,
I,
don't
know
we
could,
even
in
impute
values,
for
areas
where
we
have.
You
know
some
ambiguity,
so
I
mean
that's
not
entirely.
Like
I
mean
it's,
it's
a
myth.
It's
a
technique.
A
A
A
What
criterion
for
how
constant
size
parameters
for
given
solar
across
frames,
yeah
so
I
mean
we
can
do
things
like
look
at
the
constant.
Like
you
know,
we
can
look
at
how
the
data
changeover
frames
I
mean
again.
The
algorithm
is
gonna
pick
up
stuff,
you
know,
you
know
it
should
be
systematic,
though
I
mean
it
shouldn't
be
like
you
should
be
able
to
detect
whether
it's
it's
picking
up
the
right
things
or
not
pieces,
because
we're
talking
about
you
know,
model
and
we're
talking
about.
A
Like
you
know,
a
machine
learning
model,
it's
generalizing,
a
training
set
so
we're
training
it
on
these.
You
know
rectangles,
basically,
and
we're
saying
pick
something
that
looks
like
this
and
it'll
pick
the
objects
out
of
the
image
and
it'll.
Do
it
in
a
way?
That's
not
like
staring,
you
know
doesn't
have
to
be
exactly
like
it.
It
can
be
a
little
bit
different,
maybe
a
little
bit
bigger,
maybe
a
little
bit
different
shape,
but
it'll
detect
it
and
it'll
say
okay.
A
This
is
this
is
an
object,
but
the
thing
is,
it
could
pick
up
like
I.
Think
we'll
see,
maybe
in
some
of
the
names
it'll
pick
up
things
in
the
background
that
look
like
it
too,
and
so
we
can't
be
absolutely
sure
that
we're
gonna
get
what
you
showed
us
in
the
picture
that
you
put
on
your
screen
share
a
wall,
yeah
subtle
point:
if
width
of
a
cell
increases
it
may
be
dividing
okay,
yeah
there's,
and
so
there
are
a
lot
of
things
there
that
are
gonna.
A
So
should
leaders,
single
daughter
cells,
so
yeah
I
mean
they're
gonna,
be
probably
things
parts
of
the
movie
where
we're
gonna
have
a
lot
of
ambiguity
and
then
well
we'll
deal
with
that
when
we
get
to
it
because
I
think
there
might
be
some
interesting
methods
for
sort
of
you
know,
figuring
out,
you
know
what
we
can
identify.
First
of
all,
we
can
identify
outliers
for
things
that
don't
look
correct
like
if
you
have
like
dividing
cell
it
might
not
be
detected.
A
It
might
be
out
of
sequence,
and
so
then
we
can
focus
on
those
things,
and
you
know
there
might
be
ways
to
train
it
to
just
identify
those
things.
So
we
can
go
back
through
the
images
and
find
those
things
and
then
combine
the
data
together.
So
I
mean
this
is
like
a
you
know:
we're
gonna
have
to
get
maybe
get
a
little
bit
creative
in
terms
of
the
machine
learning
application
here.
A
I,
don't
know
well
we'll
see,
but
the
next
step
for
you
guys
is
just
to
like
in
it
get
some
data
and,
let's
kinda
you
know
the
first
step
of
reconstructing
the
the
organism
is
to
plot
the
data
just
to
see
if
it
makes
any
sense
at
all.
If
you
can
see
like
if
it
looks
like
the
movie,
if
there's
a
sequence
of
movement,
for
example
it
spirals
around,
and
then
you
see
that
in
the
movie,
then
you
have
a
pretty
good
chance
that
it's
picking
up
on
a
real
feature.
A
A
Coming
along,
so
that's
good,
why
don't
we
move
on
to
the
knee
in
his
project?
So
Richard
said:
if
you
have
a
rectangle
angle,
is
this
trigonometry?
So
you
can
do
there
are
a
lot
of
shortcuts
to
getting
like
angles
from
data
like
you,
you
can
use
the
motion
between
two
frames
or
three
frames
to
figure
out
like
to
find
an
angle
just
by
finding
like
you
know
whether
it
forms
some
sort
of
triangle
and
then
use
trigonometry,
but
yeah.
There
are
a
lot
of
tricks
there
we'll
talk
about
it
later,
though
so
Vinay.
A
F
A
I
mean
well,
we
talked
about
the
how
it
was
identifying
background,
blobs
his
cells
and
then
I.
We
were
talking
about
how
to
get
around
that,
and
then
I
mentioned
the
thing
about
like
finding
a
lot,
basically
determining
it
to
be
an
outlier
in
the
embryo.
So
if
you
have
a
bunch
of
cells,
Rob
is
way
out
here
you
can
use
distances
between
the
centroids
of
those
cells
to
sort.
B
F
F
C
D
E
F
E
F
F
C
A
Okay
yeah,
that
sounds
good
I,
also
wanted
to
point
out
that
coming
up
this
next
week
and
I'll
remind
you
in
the
newsletter,
we
have
a
second
evaluations
for
G
soccer,
so
I'll
be
doing
yours
and
then
I
think
you
have
a
sail
evaluation,
so
they
get
they're
getting
really
ups
they're
getting
really
kind
of
nervous
about
students
coming
too
close
to
the
deadline,
so
I
don't
know
but
anyways.
So
why
don't
we?
Actually?
Let
me
go
back
to
the
chat.
A
We
have
a
couple
more
things
going
on
here:
okay,
so
Ottawa
responded
to
Richard,
but
Sir
these
rectangles.
These
are
rectangles
in
the
image
for
analysis.
We
have
XML
files
with
some
coordinates.
We
need
to
make
sense
of
it.
It's
very
nice
visualization
right
quite
tricky
to
process.
So
that's
something
that
we
can
work
on.
A
E
A
A
Then
Richard
asked
any
of
you
start
drafting
a
manuscript
as
well
yeah.
He
did
show
the
document
and
he
hasn't
made
it
public
yet,
but
that's
going
to
be
basically
I
guess
people
are
starting
to
do
the
method,
make
sure
the
methods
are
nailed
down
and
then
I
have
yet
to
like
get
a
full
manuscript
drafted,
but
I
think
once
we
have
the
methods
you
know,
it'll
it'll
come
a
little
bit
easier
to
get
everything
organized
into
a
manuscript,
but
yeah
we've.
Actually,
we've
been
while
Richards
been
gone.
A
We've
been
discussing
that
a
little
bit
about
you
know
how
we
would
do
the
manuscript
I
thought
we'd
start
with
the
methods
first.
So
that's.
What
is
what
we're
showing
and
so
we'll
just
you
know
eventually,
you
know
well
I
want
you
guys
to
push
the
document
to
github,
we'll
just
do
it
there
and
we
can
edit
in
at
least
you
know,
parts
like
the
methods
and
then
we
can
assemble
it
in
Google
Docs
later.
But
so
why
don't
we
go
to
the
boards
now?
D
A
If
we
go
into
that
repository,
we'll
go
to
the
projects
and
then
we'll
go
to
the
board,
which
is
a
Summer
of
Code
project,
and
we
have
our
issues
here.
We
have
in
progress
and
to
do
so
I'm
progress.
We
have
a
couple
of
things.
We
have
test
output
images
with
k-means
clustering
is
that
Gunners.
That
is
I,
think
it's
probably
done.
A
A
E
C
F
D
C
A
F
A
F
C
A
A
Know
I
met
in
the
account
right
now.
This
is
the
link
to
the
repository
and
then
we
go
to
digital
bas.
Alright,
we
go
to
projects
tab
and
then
we
go
to
the
digital
vessel
area
board
and
now
we
have
our
issues
here.
We
have
in
progress
and
to
do
so,
if
one
of
you
guys
usual
as
matter
peratt's,
can
move
the
issues
around
when
I
say
move
them.
That
would
be
good.
So
we'll
start
with
in
progress,
number
37
acquire
results
for
trainings.
A
G
A
That
progress
mathematical
model
development
number
twenty
seven.
So
this
is
all
tricky
because
now
we're
talking
about
mathematical
models
for
the
the
basal
area
itself
and
we're
actually
talking
also
about
mathematical
models,
were
validating
the
results.
So
why
don't
we
leave
this
in
progress?
Maybe
you'll
think
about
this
a
bit
more?
A
He
said
you
know,
maybe
if
we
had
it,
if
we
could
get
it
outputted
from
we
have
the
XML
files,
but
maybe
get
like
output
it
as
a
CSV
file
as
well.
So
we
have
like
everything,
is
comma
delimited
and
then
from
there
you
can
import
it
into
something
like
Excel
or
into
Python,
where
you
can
actually
do
some
analysis.
A
Okay,
so
we'll
we'll
talk
about
that
later,
but
so
number
17,
then
adding
information
about
the
biology,
Abascal
area,
we're
still
that's
kind
of
still
open
description
or
neural
networks,
approach
for
biologists,
I
think,
that's
probably
what
we
were
getting
at
with
the
methods.
So
that's
in
progress
to
number
31
template
methods.
Number
29
oswald
was
showing
us
his
methods
that
he's
working
out
and
so
that'll
be.
You
know,
you'll
use
that
as
a
template
for
all
your
methods,
you'll
kind
of
like
there
will
be
a
lot
of
repetition
in
terms
of
the
format.
A
So
you
know
you
will
have
a
section
for
each
Odle
piece
of
detail
that
you
do
so.
If
you
do
something
in,
if
you
do
some
apply
some
technique,
you
want
to
write
a
description
of
what
you
did
maybe
a
little
bit
about
how
it
works,
because
we're
trying
to
appeal
to
a
broader
audience
than
just
machine
learning
people
or
so
we
want
to
have
a
little
bit
of
description
in
there.
But
I'll
work
with
you
on
the
editing
that
text
as
well
so
and
then
number
30
is.
D
A
A
So
you
know,
don't
don't
think
you
have
to
like
nothing.
You
have
to
master
the
paper.
Just
think
just
kind
of
look
at
the
methods
section
and
that's
I
mean
it
seems
like
that's
what
you're
doing
well
we'll
talk
about
that
more
later
issue
number
35
is
labeling
data
set.
So
this
is
I
think
this
is
done.
Isn't
it
yes?
A
A
You
know,
building
a
bivariate
graph
of
a
of
the
points
that
are
generated
by
your
output,
and
so
that's
that's
in
progress,
of
course,
but
we'll
probably
create
some
issues
around
that,
because
right
now
it's
still
like
kind
of
vague
is
how
we're
gonna
do
that
yeah,
but
I
think
that's
I
think
that's
a
good
way
to
way
kind
of
get
a
handle
on
what
we
have
what
we
can
get
and
then
you
know
reconstruction
is
not
easy.
Just
let
you
know
right
now.
A
If
I
take
a
couple
passes
different,
like
maybe
use
a
couple,
different
training
sets
to
actually
get
a
while.
You
know
all
the
information
out
so
but
I
think
we're
we're
headed
in
the
right
direction.
So
to
do
is
we
have
a
couple
of
issues.
One
is
model
features
with
him
in
between
frames,
and
then
we
kind
of
talked
about
that.
A
little
bit
so
far.
Leave
that
to
do,
and
we
make
a
note
about
number
28.
A
We
might
create
a
couple
of
issues
around
this
as
well.
Sort
are
around
7
and
28
because
we're
talking
now
about
like
getting
multiple
frames
and
tracking
between
frames
and
so
I
think
we
maybe
we
have
a
little
bit
better
idea.
What
that's
gonna
look
like
create
video
resources
for
basilar
area
biology,
that's
open,
I,
don't
know!
If
anyone
has
any
ideas
about
that,
but
you
can
you
know
it's.
It's.
C
A
If
anyone
has
any
ideas,
we'll
discuss
them
and
then
define
features
within
in-between
frames,
so
actually
3
and
28
are
linked,
I
think
they're.
Finding
the
features
are
then
in
between
frames
might
be
a
case
of
where
we
have
the
data,
and
then
we
plot
the
data
out,
maybe
a
graph
or
you
know
we
model
in
some
way.
We
actually
find
that
there
certain
features
that
you
know
like
maybe
characteristic
movements
or
curves.
A
But
that's
you
know
that's
after
we
get
the
data
out
and
we
can
really
analyze
it
and,
like
I
said
I
want
to
be
in
the
loop
on
like
once
we
get
the
data
out,
they
will
help
you
go
through.
In
fact,
we
might
actually
do
in
a
meeting
where
we
kind
of
walk
through
the
data
a
little
bit,
and
you
know
plot
things
out
and
like
we
can
really
get
a
handle
on.
A
What's
there
so
I
think,
that's
something
that
maybe
we'll
do
you
know
me
in
the
next
few
weeks
whenever
we're
ready
for
that,
whatever
we
can
have
a
data
set
output,
it
that's
just
numbers
and
that
you
know
represents
a
large
number
of
frames
and
you
know
we
can
actually
track
cells.
So
we
have
the
identification
number
and
everything
so
well
we'll
do
that
in
next
few
weeks.
So
anyone
have
any
other
issues
they
want
to
add
to
the
board.
I
mean
I
mentioned
that
we
were
going
to
yeah.
B
A
Mean
I
was
just
seeing
about
wake
breaking
that
out
playing
so
wicked.
It's
probably
that
issues,
probably
a
couple
of
steps,
so
I
want
to
make
sure
that
we
kind
of
like
take
it
one
step
at
a
time
and
kind
of
address
them
independently,
but
I
mean
it's
del.
So
if
you
just,
we
just
you
know,
go
ahead
and
do
it
and
then
find
that
we
address
all
these
issues
at
once.
That's
fine
too,
but
yeah
I
mean
so.
A
If
anyone
has
any
ideas
for
issues
across
over
the
course
of
the
week
you
now
you
have
the
power
to
put
them
into
the
board
as
you've
seen
your
power
to
move
issues
around
so
I
encourage
you.
If
you
have
a
good
idea
of
what
you
want
to
accomplish
with
something,
and
you
think
it
deserves
an
issue,
put
it
on
the
board,
make
it
sort
of
in
a
you
can
do
that
as
well.
So
let's
stop
sharing
that.
So,
if
that's
it
for
the
boards,
then
I
think
we're
on
track
for
this
week.
A
Good,
that's
great!
Thank
you
all
right!
Everyone!
Well,
thank
you
for
attending
the
meeting
and
I
look
forward
to
your
discussion
this
week
over
the
course
of
the
week
on
different
things.
If,
if
you
want
to
talk
and
slack
or
email,
let
me
know
if
we
want
to
get
you
know
if
we
have
an
issue
that
comes
up
or
we
have
an
idea
of
something
we
want
to
kind
of,
you
know
add
to
the
mix.
Let's
communicate
about
that.
Otherwise,
we'll
we'll
see
you
next
week.