January 21, 2025 • 24 mins
Invention is the wellspring of technology that makes innovations in 5G, AI, quantum computing, and video compression possible. Yes, video compression, the technology that enables us to take hours of 4K video on our smartphones. The technology that allows Netflix to stream 4k and eventually 8K movies and TV shows to your home. The technology that has democratized video conferencing on our devices and much more.
neXt Curve’s Leonard Lee had the opportunity to speak with master inventor, Dr. Marta Karczewicz, VP of technology and fellow at Qualcomm, about her career-long cotribution to a technology that has yielded billions if not a trillion dollars in economic value and continues to be a foundational technology enabling the advancement of digital media, gaming, XR, IoT, robotics, automotive, AI and more. Marta has over 800 patents to her name and has been recognized as one of three finalist for the 2019 lifetime achievement in invention award by the European Patent Office.
Marta & Leonard discuss the following topics in a chat with a master inventor:
➡️ Introducing the Patent Trillionaire, Dr. Marta Karczewicz (0:30)
➡️ The essentiality of video compression (2:50)
➡️ Why Marta is one of the leading inventors in the world (6:45)
➡️ Leonard upgrades Marta’s inventor status (7:49)
➡️ The foundational pillar of digital tech and applications (10:14)
➡️ Marta’s top video compression trends for the future (16:15)
Please subscribe to our podcast which will be featured on the neXt Curve YouTube Channel. Check out the audio version on BuzzSprout - https://nextcurvepodcast.buzzsprout.com or find us on your favorite Podcast platform.
Also, subscribe to the neXt Curve research portal at www.next-curve.com for the tech and industry insights that matter.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:06):
Next curve.
Hey
Leonard Lee (00:09):
everyone.
This is Leonard Lee, executiveanalyst at NextCurve.
And I am here in beautiful SanDiego.
It's a beautiful day here.
Right.
And I am joined by MartaKartevich.
Did
Marta Karczewicz (00:24):
I get that right?
Yes, you got it right.
And she,
Leonard Lee (00:27):
she is the vice president of technology at
Qualcomm and she is also.
A fellow, and I just wanteveryone to know she is the most
humble trillionaire that you'llever meet.
Very, very humble.
And we're going to try to gether to talk about her
(00:49):
technology, her background, andsome of the revolutionary things
that she's been involved withand has enabled through her
career.
Are you going to be able tohandle this?
Marta Karczewicz (00:59):
Yeah.
Leonard Lee (01:00):
You know, cause last time we spoke, you were
way, way too humble, and, but
Marta Karczewicz (01:06):
good.
Leonard Lee (01:06):
it is.
And, I think this is going to bea great story for you to share
with the next curve audience,because, we talk about
innovation all the time, right?
we overuse that term, and wealso probably under appreciate
invention.
And, yes, we have F 35s flyingover us, we're trying to do
(01:29):
this, so you'll have to bearwith us.
But invention is also, I think,something that's probably under
appreciated.
Maybe not overused, but underappreciated.
So, I really wanted to talk toyou about that, especially in
the context of the research andthe work that you've been doing
at Qualcomm as it relates tocompression technology, right?
(01:53):
And so maybe talk about yourbackground a little bit, your
role at Qualcomm.
Marta Karczewicz (01:57):
So, yeah, I have been working on, video
compression or other forms ofcompression, for close to 30
years now.
Wow.
started originally around, 95.
I started in Nokia, been therefor 10 years.
Then switch, briefly toMicrosoft and now, I think over
(02:20):
18 years, at Qualcomm.
all this time, I have beenworking on mostly video
compression.
when I was starting, it was avery, very fresh area.
We had just one, standardizedvideo codec and one codec in
usage.
maybe I'll just tell briefly whydo we even need video codecs.
(02:45):
I think everybody knows whatvideo is.
We keep watching it.
We have broadcast.
streaming.
zoom calls and so on.
But what people sometimes do notrealize that, after we capture
it by camera, it has to becompressed.
Otherwise, this whole ecosystemjust wouldn't work.
Right,
So just to give us some kind of, let's say,
(03:09):
background, how big videowithout compression would be.
Yeah.
let's think about what's becoming popular
these days, which is, 8K video.
Let's say 60 frames per second.
Well, that's now I would sayunderestimate because people
want to watch sport in like 120frames per second, 10 bit.
(03:32):
If you want to see the send theraw data, that would be around
over 30 gigabits per second.
I would say.
Probably almost nobody has thatlevel of Internet connection.
And let's remember
that this numbers are being increased.
So basically what the videocodec is doing is grabbing this
(03:55):
raw, raw data.
Yeah.
And trying to, reduce the size without reducing
visual quality.
And,
this day's video codecs compress video.
hundreds or even thousands oftimes.
Before, you get it on your end.
Yeah.
that's what I have been devoting my life to
(04:16):
the Kodak from the times that Istarted, which I said the only
Kodak, at, on the market was pecto, evolve quite a bit, just to
give, some kind of, Comparisonpoint at the codec that we were
working now, can compress thedata that MPEG 2 was able to
(04:37):
compress because MPEG 2 is notable anymore to do everything
that we are compressing thesedays.
That's not going to happen.
But if we take the content thatMPEG 2 still could do and
compare it with what we havenow, we are using less than 15
percent of what MPEG needed.
to get the same visual quality.
(04:58):
Oh, yeah.
the top of everything, MPEG 2, couldn't do,
what we keep seeing noweverywhere, which is this high
dynamic range and, couldn'texactly do, what we are using in
teams, which is compressinggraphics and text.
I mean, it could do it, but thebitrate would.
Probably explode.
(05:19):
You might as well not compressit at all.
I would like to think that I atleast partially contributed to,
this experiment.
Leonard Lee (05:27):
Too modest.
Marta Karczewicz (05:29):
Improvement.
So I have been workingpersonally on all the generation
of the codecs after, MPEG 2.
AVC, HEVC, VVC.
And now, the new codec that weare starting, which, hopefully
will be H.
267, I also have quite a fewpatents in this area, but last
(05:51):
time that I checked, I think Ihad over, 800 granted patent in
US and 250 in Europe.
Leonard Lee (05:58):
Wow, did you hear that?
That's pretty amazing.
Marta Karczewicz (06:01):
And I've been actually probably mainly
recognized for the patents.
I have been one of the threenominees of Lifetime Achievement
Award by European Patent Officein 2019.
And I have some IP, awards.
Also from Qualcomm, one of 2012and so on and so forth.
(06:27):
This is like the patents and Iwould say this is like the side
product of my work.
Sure.
so that's roughly what I do and who I am.
Leonard Lee (06:35):
Yeah so they call you the patent billionaire,
right?
And I actually consider you moreof a trillionaire because I'll
tell you why.
I was one, I think I might'vebeen one of the first.
I consumers to buy one of those,DV camcorders.
Yes.
Right back in the day.
And when I transferred the miniDV, one hour of mini DV footage
(06:59):
into a digital format.
Of course, this is 480presolution.
It took up 25 gigs of storage.
And so, we look at thetechnology that you've helped to
develop and continue to doresearch in has had on just
(07:21):
everything that's digital,because keep in mind mobile,
wireless traffic, 70 percent ofhis video, like you're saying
media, think about outside ofthat internet traffic, probably
the same order, of 70% is, ifnot more, is probably all video,
right?
And so imagine a world where weare still at four 80 p with the,
(07:43):
video quality at 25 gigs.
We would not have what we havetoday in terms of the.
The quality of video and mediaexperiences, right?
Like you mentioned, broadcast.
Marta Karczewicz (07:58):
with the resolution.
Why would anybody buy 80 inch TVif you Yeah.
Exactly.
So you see big blocks ofidentical color.
I was like, what's the point?
Leonard Lee (08:11):
Right, right.
Now, you know, going into thefuture as we look at higher
resolutions, new, innovativeMedia formats, you know, I can
only imagine you have vision onwhat that looks like, right?
And, and, you know, again, likeyou mentioned the experience,
these are the higher levelexperiences, higher quality
experiences that we've been ableto achieve, through a smaller
(08:36):
data footprint.
And that has enabled so manythings, whether it's broadcast,
whether it's, streaming, allthese different modalities and
ways of transmitting,distributing and storing even,
right?
Marta Karczewicz (08:52):
would say, yeah, yeah., because I
mentioned, I use the word whatMPEG 2.
still could do.
And when you look how the videoevolved, it's not that we can,
compress what MPEG couldcompress, like almost 10 times
better.
But it's also that we enable alot of things that MPEG do could
(09:12):
not, do.
Like I mentioned that.
You go to any shop and TV likelyis advertising high dynamic
range with color gamut, and it'strue.
It, it made a big difference inthe watching experience.
Yeah.
Like the older video, like the maximum
brightness that we see was likearound 100.
(09:33):
what we usually see as humanscan go up to 10,000 current TVs
are being advertised at two oreven 3000 needs.
So basically already thebrightness that you see, which
can be captured in display ismuch, much higher.
contrast ratio, that's anotherthing that, before we may be
(09:53):
wow.
Were able to compress somethingthat was.
for like two to 10, uh, thesedays, that goes much higher.
And what humans can do is likemaybe two to power of 20 levels
of Congress.
And we have, I'm all at thisplace, but their camera, so we
can show it.
color reproduction also at thetime was rather very far from
(10:17):
what we can do now, because theyare addressing this wider color
gamut.
So it's couple of times betternow.
Basically, the colors that wecan reproduce and, the, the
codex, especially starting fromHVC were able to address it so
we can compress this formats.
we also can, compress thingsthat became important while the
(10:41):
usage of video was evolving,like, for example, during Corona
virus almost everybody switchedto.
online experience and thatincluded work and the zoom call
and team calls, with sharing theslides, spreadsheet, or even,
I'm still programming.
So even showing like the sourcecode via video, that became like
(11:06):
became part of the normal.
Experience.
Yeah.
And, past codex like did notreally assume that you will
compress anything but thenatural contact.
And the newer codecs like, HEVCAnnex and VENVVC can do that.
So basically, and they can do ittwice better than the past
(11:26):
codec, even if you look at allother tools.
So we have the evolution and,what I see at least the
evolution slowly going is,again, enabling, more formats
because they become, more widelyused.
Yeah.
one of the examples is gaming.
(11:47):
Let's start from entertainmentactually still because what
people do not often realize thatall this multiple services that
pop up with online gaming.
Yeah.
using the codecs, actually video codecs,
so basically GPU produces thecontent, a codecs grabs it,
sends it to the end user, whenyou react to it, the reaction
(12:10):
goes back, and then basically,so on and so forth, one of the
Trends that we see now is thequestion.
how do you use, the GPUinformation to do better and
coding because, what we, forexample, deal with in normal
codex is, like if you thinkabout video, it's really just,
(12:33):
collection of still images.
A lot of redundancy is that,From one image to the other.
They are mostly similar.
There is just an ocean of someobjects, but it's not that easy
to estimate.
Well, when you have GPU, you canget information per pixel,
however pixel move.
Yeah.
So, okay.
What new technologies can you doto do it?
(12:55):
Another thing which is now veryhot and also somewhat relies on
video or maybe more than peopleeven imagine is everybody talks
about, AI, and video coding formachines,
which
basically means that, we have all this
cameras, we have cameras, notonly, like security cameras, but
there are, cameras in factories,warehouses.
(13:18):
wherever something can happenand some intervention might be
needed.
So, okay, some of this contentcan be analyzed directly in
camera, but usually it'sactually being sent to the edge.
There
it's being analyzed by AI on the server.
And how it's being sent asvideo.
(13:41):
So basically, now what we werelooking more and more is that
what we perceive as importantas, humans when watching, it's
not that important for AI veryoften.
So basically, how do we changecompression to make it, better
for, AI?
Leonard Lee (13:59):
Beyond that, what's that one other thing that really
has you excited, that you'rereally looking forward to, that
you
Marta Karczewicz (14:05):
think
Leonard Lee (14:05):
compression technology is going to have a
revolutionary impact on?
Marta Karczewicz (14:10):
I have one thing that is exciting to me
from a technical point of view,which is usage of AI in codecs.
Not only that it's being, youknow, later used by AI, but how
to introduce, neural networksinto video codecs.
That's already starting in thestandards that we are working,
(14:30):
that's already being tested.
It's probably going to evolve atsome point that the whole codec
might be, neural network based.
Yeah.
But it's not going to happenovernight, to be honest, since,
the main issue is thecomplexity.
I mean, the codecs are designedthat encoder can be very
complex.
Right.
But the decoder has to be very
(14:51):
lightweight because it's runningon, for example, mobile device
and you don't want it to killyour battery in one hour or
less.
Neural networks are moresymmetric in their design, so
we'll have to figure out what todo with it, but they open new
opportunities.
I see mainly like two.
(15:14):
I mentioned this video codingfor machines that the quality
measure will have to change,
Mm-hmm
normal product, let's say have the very
fi they have very fixed design.
If you wanna change, the waythat, you measure quality, you
have to redo the whole encoder.
Parts of the theod are verycomplex.
With neural network, actually,you can just retrain the weights
(15:37):
of the codec for a differentquality measure.
so that would make it, muchmore, user friendly when you
change the scenario where codec
is used.
We have been experimenting, literally, for
years, how do we compresstexture.
Because, if you look, forexample, around, you have tons
(15:59):
of trees, right?
the same if you look at, water,everything that has texture, the
codecs didn't really progressthat much.
How to compress it, because it'srandom, it's very difficult to
find what type of prediction youshould use.
It's almost like it's a noise.
Neural networks, they wouldn'tcompress it exactly, but you
(16:23):
basically can tell them, don'tcompress it exactly, but, mimic
the statistic of this texture,mimic the statistic of this
noise.
And then when it does it, itdoes it, I would say perfectly.
You can get, almost 10 timesslower, betray when you get for
(16:46):
the normal codex.
And yes, if you, do thedifference, the difference will
be big between the original andthe current image.
But when you look at it, Youcan't tell.
It looks fantastically natural.
Leonard Lee (17:00):
So is it like a smoothing effect?
Marta Karczewicz (17:02):
not smoothing.
It just has a different error,measure.
It's just like, these arestatistical properties and these
are the statistical properties.
Try to match them.
neural network will do it.
So
yeah,
because normal codecs, you are right.
Short of like this graphiccontent, they rely on assumption
(17:23):
that, the content is smooth withsome edges in between.
Right.
the motion is fairly, predictable.
Well, again, like I'm looking atthree in front of me.
it's very textured and themotion is all over the place
because the leaves are moving.
Which other way?
So that's one thing that I wouldsay excites me.
(17:43):
And then other thing whichbecomes, that might be a bit
more longer term is theimmersive video.
Okay.
So we are used to, just watching 2D video.
There was some excitement about3D video, which kind of came
back a bit.
I would say, I don't want tocall it a fake 3D video, but it
(18:05):
is, because what we usuallyactually do, we just capture
videos from two differentviewpoints.
Yeah.
So it's a stereoscopic videoand, it looks 3D, but, for
example, like when we move ourhead, we see that objects move
versus each other.
when we watch this.
3D videos.
(18:25):
Nothing is moving.
So basically I was like, itlooks 3D when I don't move.
But when I move, it starts tolook fake.
It's more like volume.
It's missing the volumetricaspect.
So basically
we are working on multiple formats to be able
to support.
(18:45):
volumetric video
so
that it will not only allow you like to some
extent is just to allow morerealistic 3D experience, but,
there are also attempts to makeit truly immersive, not maybe
for everything, but for example,when you look at, sports, most
of the events are captured byCameras surrounding the stadium.
(19:09):
Yeah.
And what we sometimes even watch on our TV.
It's not even a real video.
It's a video produced by,building and 3D model and
capturing it like, let's say,view that you cannot see or
camera doesn't capture, but youcan recreate it by, capturing
(19:30):
how this 3D model would lookfrom this or that point of view.
Yeah.
So now the question is, if wecan already do that, can we send
this information to the, enduser and then they can replace
being, for example, receiver orbe a quarterback or whichever
they want to be.
(19:51):
So basically, but that I seestill as a longer term
evolution.
Because that requires also, Ithink evolution of the capture
is maybe in some extent, to someextent there, but it Requires
evolution of the display.
To some extent you probably canalready do it with XR.
(20:12):
but that still will take timebefore it will be, that popular
or that, broadly use that,production of such content will
make, Sense from the cost pointof view.
So I would say it's like,Enhanced and, immersive.
That's what really, reallyexcites me at the moment.
Leonard Lee (20:32):
Well, you know what really excites me?
That I learned so much just inthe short conversation.
And I love that you told me thatI was wrong.
See, I got schooled on my ownpodcast.
It's like, but you know, I'mtalking to a, trillion dollar
inventor here.
That is just, oh
Marta Karczewicz (20:51):
yeah.
There's just so many
Leonard Lee (20:53):
patents.
It's like, but no, you know, oneof the key takeaways I've had,
you know, I've already spoken toyou about.
Your background and many of thetechnologies that you're working
on have worked on in the past.
And I'm impressed by the impact.
It's not like hypotheticalimpact.
It's like real, real impact.
That's had just an immense haloeffect.
(21:15):
Is also now that we've spokenagain, really that foundational
technology that enables,experiences to continue to
evolve and advanced.
And without that, we're stuck,right?
And then the innovation acrosswhether it's a streaming
broadcast or capture andconsumption, all that stuff just
(21:36):
stand still.
when people think of Qualcomm,they think of Snapdragon now.
They used to think of Qualcommfrom a, a wireless 5G
perspective.
But then there's this otherthing that, Qualcomm is doing
through the work that you'redoing that, is really another
one of those foundationalpillars, so that's the other
(21:56):
thing that I think, we'resharing with the NextCurve
audience that they might notknow.
I mean, I don't think a lot ofpeople know that you got this.
Most of the
Marta Karczewicz (22:02):
codecs that we talked about are implemented on
Snapdragon.
Leonard Lee (22:06):
Come on.
Marta Karczewicz (22:07):
Oh, okay.
We have to educate
Leonard Lee (22:09):
The consumer doesn't know.
A lot of people don't know aboutthis stuff.
You know, they just see a bunchof acronyms in the spec, you
know, technical spec sheet.
And then, but no, I, I reallyappreciate your taking the time
out.
of your busy day and, sufferingthe F 35s flying over him.
By the way, speaking ofcompression, a turbofan engine
(22:30):
uses compression, it compressesair.
That's what makes, all thesefighter planes go supersonic.
Just like what she's doing, youknow, she's doing compression
that makes our videos go.
Super what?
Visual?
Yes, but you
Marta Karczewicz (22:45):
can also think about it that way.
That latest Top Gun moviewouldn't be as cool without the
codex that we make.
Leonard Lee (22:55):
Oh my God.
Wow.
She is good.
She is good.
Well, Marta, thank you so much.
I really appreciate the time andthe education, quite honestly.
And, I hope that we can continueto conversate every single time.
I'm serious.
Every single time I sit down andtalk to you, light bulbs go off
in my head because, I do broadTMT, like technology, media, and
(23:15):
telecom research.
And so as you were speaking, itwas just I saw the future, I saw
the future and it's absolutelywonderful.
So once again, thank you.
And, to, our audience, thank youfor tuning in.
We hope you enjoyed this,discussion and hopefully you
learned as much as I did, aboutthis wonderful individual who is
(23:37):
the most humble, revolutionary,a disruptor.
I could even call her adisruptor.
She deserves that titledisruptor.
And of course, in all thepositive and wonderful ways,
follow us on next curve, WWW.
Dot next dash curve dot com.
And remember to follow, like,share, the content on our
(23:59):
YouTube channel and we will seeyou next time.
And once again, thank you somuch.
Next curve.
Hey
Leonard Lee (00:09):
everyone.
This is Leonard Lee, executiveanalyst at NextCurve.
And I am here in beautiful SanDiego.
It's a beautiful day here.
Right.
And I am joined by MartaKartevich.
Did
Marta Karczewicz (00:24):
I get that right?
Yes, you got it right.
And she,
Leonard Lee (00:27):
she is the vice president of technology at
Qualcomm and she is also.
A fellow, and I just wanteveryone to know she is the most
humble trillionaire that you'llever meet.
Very, very humble.
And we're going to try to gether to talk about her
(00:49):
technology, her background, andsome of the revolutionary things
that she's been involved withand has enabled through her
career.
Are you going to be able tohandle this?
Marta Karczewicz (00:59):
Yeah.
Leonard Lee (01:00):
You know, cause last time we spoke, you were
way, way too humble, and, but
Marta Karczewicz (01:06):
good.
Leonard Lee (01:06):
it is.
And, I think this is going to bea great story for you to share
with the next curve audience,because, we talk about
innovation all the time, right?
we overuse that term, and wealso probably under appreciate
invention.
And, yes, we have F 35s flyingover us, we're trying to do
(01:29):
this, so you'll have to bearwith us.
But invention is also, I think,something that's probably under
appreciated.
Maybe not overused, but underappreciated.
So, I really wanted to talk toyou about that, especially in
the context of the research andthe work that you've been doing
at Qualcomm as it relates tocompression technology, right?
(01:53):
And so maybe talk about yourbackground a little bit, your
role at Qualcomm.
Marta Karczewicz (01:57):
So, yeah, I have been working on, video
compression or other forms ofcompression, for close to 30
years now.
Wow.
started originally around, 95.
I started in Nokia, been therefor 10 years.
Then switch, briefly toMicrosoft and now, I think over
(02:20):
18 years, at Qualcomm.
all this time, I have beenworking on mostly video
compression.
when I was starting, it was avery, very fresh area.
We had just one, standardizedvideo codec and one codec in
usage.
maybe I'll just tell briefly whydo we even need video codecs.
(02:45):
I think everybody knows whatvideo is.
We keep watching it.
We have broadcast.
streaming.
zoom calls and so on.
But what people sometimes do notrealize that, after we capture
it by camera, it has to becompressed.
Otherwise, this whole ecosystemjust wouldn't work.
Right,
So just to give us some kind of, let's say,
(03:09):
background, how big videowithout compression would be.
Yeah.
let's think about what's becoming popular
these days, which is, 8K video.
Let's say 60 frames per second.
Well, that's now I would sayunderestimate because people
want to watch sport in like 120frames per second, 10 bit.
(03:32):
If you want to see the send theraw data, that would be around
over 30 gigabits per second.
I would say.
Probably almost nobody has thatlevel of Internet connection.
And let's remember
that this numbers are being increased.
So basically what the videocodec is doing is grabbing this
(03:55):
raw, raw data.
Yeah.
And trying to, reduce the size without reducing
visual quality.
And,
this day's video codecs compress video.
hundreds or even thousands oftimes.
Before, you get it on your end.
Yeah.
that's what I have been devoting my life to
(04:16):
the Kodak from the times that Istarted, which I said the only
Kodak, at, on the market was pecto, evolve quite a bit, just to
give, some kind of, Comparisonpoint at the codec that we were
working now, can compress thedata that MPEG 2 was able to
(04:37):
compress because MPEG 2 is notable anymore to do everything
that we are compressing thesedays.
That's not going to happen.
But if we take the content thatMPEG 2 still could do and
compare it with what we havenow, we are using less than 15
percent of what MPEG needed.
to get the same visual quality.
(04:58):
Oh, yeah.
the top of everything, MPEG 2, couldn't do,
what we keep seeing noweverywhere, which is this high
dynamic range and, couldn'texactly do, what we are using in
teams, which is compressinggraphics and text.
I mean, it could do it, but thebitrate would.
Probably explode.
(05:19):
You might as well not compressit at all.
I would like to think that I atleast partially contributed to,
this experiment.
Leonard Lee (05:27):
Too modest.
Marta Karczewicz (05:29):
Improvement.
So I have been workingpersonally on all the generation
of the codecs after, MPEG 2.
AVC, HEVC, VVC.
And now, the new codec that weare starting, which, hopefully
will be H.
267, I also have quite a fewpatents in this area, but last
(05:51):
time that I checked, I think Ihad over, 800 granted patent in
US and 250 in Europe.
Leonard Lee (05:58):
Wow, did you hear that?
That's pretty amazing.
Marta Karczewicz (06:01):
And I've been actually probably mainly
recognized for the patents.
I have been one of the threenominees of Lifetime Achievement
Award by European Patent Officein 2019.
And I have some IP, awards.
Also from Qualcomm, one of 2012and so on and so forth.
(06:27):
This is like the patents and Iwould say this is like the side
product of my work.
Sure.
so that's roughly what I do and who I am.
Leonard Lee (06:35):
Yeah so they call you the patent billionaire,
right?
And I actually consider you moreof a trillionaire because I'll
tell you why.
I was one, I think I might'vebeen one of the first.
I consumers to buy one of those,DV camcorders.
Yes.
Right back in the day.
And when I transferred the miniDV, one hour of mini DV footage
(06:59):
into a digital format.
Of course, this is 480presolution.
It took up 25 gigs of storage.
And so, we look at thetechnology that you've helped to
develop and continue to doresearch in has had on just
(07:21):
everything that's digital,because keep in mind mobile,
wireless traffic, 70 percent ofhis video, like you're saying
media, think about outside ofthat internet traffic, probably
the same order, of 70% is, ifnot more, is probably all video,
right?
And so imagine a world where weare still at four 80 p with the,
(07:43):
video quality at 25 gigs.
We would not have what we havetoday in terms of the.
The quality of video and mediaexperiences, right?
Like you mentioned, broadcast.
Marta Karczewicz (07:58):
with the resolution.
Why would anybody buy 80 inch TVif you Yeah.
Exactly.
So you see big blocks ofidentical color.
I was like, what's the point?
Leonard Lee (08:11):
Right, right.
Now, you know, going into thefuture as we look at higher
resolutions, new, innovativeMedia formats, you know, I can
only imagine you have vision onwhat that looks like, right?
And, and, you know, again, likeyou mentioned the experience,
these are the higher levelexperiences, higher quality
experiences that we've been ableto achieve, through a smaller
(08:36):
data footprint.
And that has enabled so manythings, whether it's broadcast,
whether it's, streaming, allthese different modalities and
ways of transmitting,distributing and storing even,
right?
Marta Karczewicz (08:52):
would say, yeah, yeah., because I
mentioned, I use the word whatMPEG 2.
still could do.
And when you look how the videoevolved, it's not that we can,
compress what MPEG couldcompress, like almost 10 times
better.
But it's also that we enable alot of things that MPEG do could
(09:12):
not, do.
Like I mentioned that.
You go to any shop and TV likelyis advertising high dynamic
range with color gamut, and it'strue.
It, it made a big difference inthe watching experience.
Yeah.
Like the older video, like the maximum
brightness that we see was likearound 100.
(09:33):
what we usually see as humanscan go up to 10,000 current TVs
are being advertised at two oreven 3000 needs.
So basically already thebrightness that you see, which
can be captured in display ismuch, much higher.
contrast ratio, that's anotherthing that, before we may be
(09:53):
wow.
Were able to compress somethingthat was.
for like two to 10, uh, thesedays, that goes much higher.
And what humans can do is likemaybe two to power of 20 levels
of Congress.
And we have, I'm all at thisplace, but their camera, so we
can show it.
color reproduction also at thetime was rather very far from
(10:17):
what we can do now, because theyare addressing this wider color
gamut.
So it's couple of times betternow.
Basically, the colors that wecan reproduce and, the, the
codex, especially starting fromHVC were able to address it so
we can compress this formats.
we also can, compress thingsthat became important while the
(10:41):
usage of video was evolving,like, for example, during Corona
virus almost everybody switchedto.
online experience and thatincluded work and the zoom call
and team calls, with sharing theslides, spreadsheet, or even,
I'm still programming.
So even showing like the sourcecode via video, that became like
(11:06):
became part of the normal.
Experience.
Yeah.
And, past codex like did notreally assume that you will
compress anything but thenatural contact.
And the newer codecs like, HEVCAnnex and VENVVC can do that.
So basically, and they can do ittwice better than the past
(11:26):
codec, even if you look at allother tools.
So we have the evolution and,what I see at least the
evolution slowly going is,again, enabling, more formats
because they become, more widelyused.
Yeah.
one of the examples is gaming.
(11:47):
Let's start from entertainmentactually still because what
people do not often realize thatall this multiple services that
pop up with online gaming.
Yeah.
using the codecs, actually video codecs,
so basically GPU produces thecontent, a codecs grabs it,
sends it to the end user, whenyou react to it, the reaction
(12:10):
goes back, and then basically,so on and so forth, one of the
Trends that we see now is thequestion.
how do you use, the GPUinformation to do better and
coding because, what we, forexample, deal with in normal
codex is, like if you thinkabout video, it's really just,
(12:33):
collection of still images.
A lot of redundancy is that,From one image to the other.
They are mostly similar.
There is just an ocean of someobjects, but it's not that easy
to estimate.
Well, when you have GPU, you canget information per pixel,
however pixel move.
Yeah.
So, okay.
What new technologies can you doto do it?
(12:55):
Another thing which is now veryhot and also somewhat relies on
video or maybe more than peopleeven imagine is everybody talks
about, AI, and video coding formachines,
which
basically means that, we have all this
cameras, we have cameras, notonly, like security cameras, but
there are, cameras in factories,warehouses.
(13:18):
wherever something can happenand some intervention might be
needed.
So, okay, some of this contentcan be analyzed directly in
camera, but usually it'sactually being sent to the edge.
There
it's being analyzed by AI on the server.
And how it's being sent asvideo.
(13:41):
So basically, now what we werelooking more and more is that
what we perceive as importantas, humans when watching, it's
not that important for AI veryoften.
So basically, how do we changecompression to make it, better
for, AI?
Leonard Lee (13:59):
Beyond that, what's that one other thing that really
has you excited, that you'rereally looking forward to, that
you
Marta Karczewicz (14:05):
think
Leonard Lee (14:05):
compression technology is going to have a
revolutionary impact on?
Marta Karczewicz (14:10):
I have one thing that is exciting to me
from a technical point of view,which is usage of AI in codecs.
Not only that it's being, youknow, later used by AI, but how
to introduce, neural networksinto video codecs.
That's already starting in thestandards that we are working,
(14:30):
that's already being tested.
It's probably going to evolve atsome point that the whole codec
might be, neural network based.
Yeah.
But it's not going to happenovernight, to be honest, since,
the main issue is thecomplexity.
I mean, the codecs are designedthat encoder can be very
complex.
Right.
But the decoder has to be very
(14:51):
lightweight because it's runningon, for example, mobile device
and you don't want it to killyour battery in one hour or
less.
Neural networks are moresymmetric in their design, so
we'll have to figure out what todo with it, but they open new
opportunities.
I see mainly like two.
(15:14):
I mentioned this video codingfor machines that the quality
measure will have to change,
Mm-hmm
normal product, let's say have the very
fi they have very fixed design.
If you wanna change, the waythat, you measure quality, you
have to redo the whole encoder.
Parts of the theod are verycomplex.
With neural network, actually,you can just retrain the weights
(15:37):
of the codec for a differentquality measure.
so that would make it, muchmore, user friendly when you
change the scenario where codec
is used.
We have been experimenting, literally, for
years, how do we compresstexture.
Because, if you look, forexample, around, you have tons
(15:59):
of trees, right?
the same if you look at, water,everything that has texture, the
codecs didn't really progressthat much.
How to compress it, because it'srandom, it's very difficult to
find what type of prediction youshould use.
It's almost like it's a noise.
Neural networks, they wouldn'tcompress it exactly, but you
(16:23):
basically can tell them, don'tcompress it exactly, but, mimic
the statistic of this texture,mimic the statistic of this
noise.
And then when it does it, itdoes it, I would say perfectly.
You can get, almost 10 timesslower, betray when you get for
(16:46):
the normal codex.
And yes, if you, do thedifference, the difference will
be big between the original andthe current image.
But when you look at it, Youcan't tell.
It looks fantastically natural.
Leonard Lee (17:00):
So is it like a smoothing effect?
Marta Karczewicz (17:02):
not smoothing.
It just has a different error,measure.
It's just like, these arestatistical properties and these
are the statistical properties.
Try to match them.
neural network will do it.
So
yeah,
because normal codecs, you are right.
Short of like this graphiccontent, they rely on assumption
(17:23):
that, the content is smooth withsome edges in between.
Right.
the motion is fairly, predictable.
Well, again, like I'm looking atthree in front of me.
it's very textured and themotion is all over the place
because the leaves are moving.
Which other way?
So that's one thing that I wouldsay excites me.
(17:43):
And then other thing whichbecomes, that might be a bit
more longer term is theimmersive video.
Okay.
So we are used to, just watching 2D video.
There was some excitement about3D video, which kind of came
back a bit.
I would say, I don't want tocall it a fake 3D video, but it
(18:05):
is, because what we usuallyactually do, we just capture
videos from two differentviewpoints.
Yeah.
So it's a stereoscopic videoand, it looks 3D, but, for
example, like when we move ourhead, we see that objects move
versus each other.
when we watch this.
3D videos.
(18:25):
Nothing is moving.
So basically I was like, itlooks 3D when I don't move.
But when I move, it starts tolook fake.
It's more like volume.
It's missing the volumetricaspect.
So basically
we are working on multiple formats to be able
to support.
(18:45):
volumetric video
so
that it will not only allow you like to some
extent is just to allow morerealistic 3D experience, but,
there are also attempts to makeit truly immersive, not maybe
for everything, but for example,when you look at, sports, most
of the events are captured byCameras surrounding the stadium.
(19:09):
Yeah.
And what we sometimes even watch on our TV.
It's not even a real video.
It's a video produced by,building and 3D model and
capturing it like, let's say,view that you cannot see or
camera doesn't capture, but youcan recreate it by, capturing
(19:30):
how this 3D model would lookfrom this or that point of view.
Yeah.
So now the question is, if wecan already do that, can we send
this information to the, enduser and then they can replace
being, for example, receiver orbe a quarterback or whichever
they want to be.
(19:51):
So basically, but that I seestill as a longer term
evolution.
Because that requires also, Ithink evolution of the capture
is maybe in some extent, to someextent there, but it Requires
evolution of the display.
To some extent you probably canalready do it with XR.
(20:12):
but that still will take timebefore it will be, that popular
or that, broadly use that,production of such content will
make, Sense from the cost pointof view.
So I would say it's like,Enhanced and, immersive.
That's what really, reallyexcites me at the moment.
Leonard Lee (20:32):
Well, you know what really excites me?
That I learned so much just inthe short conversation.
And I love that you told me thatI was wrong.
See, I got schooled on my ownpodcast.
It's like, but you know, I'mtalking to a, trillion dollar
inventor here.
That is just, oh
Marta Karczewicz (20:51):
yeah.
There's just so many
Leonard Lee (20:53):
patents.
It's like, but no, you know, oneof the key takeaways I've had,
you know, I've already spoken toyou about.
Your background and many of thetechnologies that you're working
on have worked on in the past.
And I'm impressed by the impact.
It's not like hypotheticalimpact.
It's like real, real impact.
That's had just an immense haloeffect.
(21:15):
Is also now that we've spokenagain, really that foundational
technology that enables,experiences to continue to
evolve and advanced.
And without that, we're stuck,right?
And then the innovation acrosswhether it's a streaming
broadcast or capture andconsumption, all that stuff just
(21:36):
stand still.
when people think of Qualcomm,they think of Snapdragon now.
They used to think of Qualcommfrom a, a wireless 5G
perspective.
But then there's this otherthing that, Qualcomm is doing
through the work that you'redoing that, is really another
one of those foundationalpillars, so that's the other
(21:56):
thing that I think, we'resharing with the NextCurve
audience that they might notknow.
I mean, I don't think a lot ofpeople know that you got this.
Most of the
Marta Karczewicz (22:02):
codecs that we talked about are implemented on
Snapdragon.
Leonard Lee (22:06):
Come on.
Marta Karczewicz (22:07):
Oh, okay.
We have to educate
Leonard Lee (22:09):
The consumer doesn't know.
A lot of people don't know aboutthis stuff.
You know, they just see a bunchof acronyms in the spec, you
know, technical spec sheet.
And then, but no, I, I reallyappreciate your taking the time
out.
of your busy day and, sufferingthe F 35s flying over him.
By the way, speaking ofcompression, a turbofan engine
(22:30):
uses compression, it compressesair.
That's what makes, all thesefighter planes go supersonic.
Just like what she's doing, youknow, she's doing compression
that makes our videos go.
Super what?
Visual?
Yes, but you
Marta Karczewicz (22:45):
can also think about it that way.
That latest Top Gun moviewouldn't be as cool without the
codex that we make.
Leonard Lee (22:55):
Oh my God.
Wow.
She is good.
She is good.
Well, Marta, thank you so much.
I really appreciate the time andthe education, quite honestly.
And, I hope that we can continueto conversate every single time.
I'm serious.
Every single time I sit down andtalk to you, light bulbs go off
in my head because, I do broadTMT, like technology, media, and
(23:15):
telecom research.
And so as you were speaking, itwas just I saw the future, I saw
the future and it's absolutelywonderful.
So once again, thank you.
And, to, our audience, thank youfor tuning in.
We hope you enjoyed this,discussion and hopefully you
learned as much as I did, aboutthis wonderful individual who is
(23:37):
the most humble, revolutionary,a disruptor.
I could even call her adisruptor.
She deserves that titledisruptor.
And of course, in all thepositive and wonderful ways,
follow us on next curve, WWW.
Dot next dash curve dot com.
And remember to follow, like,share, the content on our
(23:59):
YouTube channel and we will seeyou next time.
And once again, thank you somuch.
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com