Hi Aston,
I do remember this video from CES along with another one that was done as a single take with a walk around the booth. What I remember from that other video was you could watch the “driver” and see the variable delay in the OMS rendering. (watch at 1:14 2020 CES Live: Smart Eye AI-powered Eye Tracking Technology to identify driver distraction - YouTube ) Remember that, I will come back to it later.
Have you seen the BMW demo with the iConcept Ease (only a German company could come up with a name like that) where the driver could ask about what things were that they were looking at outside of the vehicle? How about the recent Cerence video, where the driver asks “Hey, Mercedes what is that building?” as she drives along. https://analyticsread.com/2021/01/cerence-look-adds-visual-ai-to-automotive-virtual-assistants/ Those systems need multiple systems with known levels of accuracy and timing to work together. Cerence does the listening, another company does the detailed mapping including I presume height information, so it can identify what is visible from the road above and behind other items, whilst not doing Xray vision through buildings. For this to work you need accurate positioning on the map for the vehicle at the time the question is asked, with the gaze vectors at the time the question is asked from the location of the eyes at the time the question was asked. Note there is a duration for asking the question and a delay in interpretation, so have to go back to the actual time for the locations. Remember, I mentioned the delay in the Smart Eye video, that is OK when identifying the driver when parked. If it is not too long it is OK while driving for DMS as long as it is consistent now that will explain why Colin gets so upset.
To have consistent timing, you need real time programming on a real time chip. ARM design regular chips like in your phone and special real time cores where the order of processing is guaranteed, so timing is quick and consistent.
HW agnostic means it wasn’t designed to run consistently, you can’t fix this in customisation.
So for a safety system - did you see the ball rolling onto the road needs constant timing to time the gaze at a moment and link to the exterior view. Same with looking at a building, those gaze vectors over time should track to a constant place on the map. If you need this level of safety AND convenience, you need realtor chips and programming, not HW agnostic software.
As you can see below, DMS algos dont have any delay
From the industry we know that performance of these algos is pretty much on the same level between DMS suppliers. For commercial success you also need to deliver and be a trustworthy companion for the OEMs. I believe this has been the strenght of Smart Eye.
Like you said that demo with lag was for OMS. I believe there are different algos for different purposes as stated by Martin in the video. I cannot say more about the delay without more knowledge of the context for that demo, but I would say that that is not for DMS purposes.
I believe there are different requirements to OMS and DMS algos.
If Smart is just china shit and still has 2/3 of DWs then competitors must be way off the scale i’ll get back to you after 2020 RfQ nominations have been published - please remember this
“I am exceptionally proud of Smart Eye’s engineering team. They have developed a scalable software solution that can run on multiple hardware platforms. This ensures the combination of optimized performance, competitive system cost and is qualified for safety critical automotive applications. This is a winning formula” says Martin Krantz, founder, and CEO of Smart Eye.
Hi Kesa, if I ask you the time, 3 times in a row, the first time you will look at your watch, the second time you will tell me the same time from memory, the third time you would sigh, then tell me the same time. That’s how modern chips work. You can’t turn that off, the delay is variable, because that is more efficient.
If you want to answer the same question with the same delay you need to use a different core in your chip and use different programming.
If Smart is just china shit and still has 2/3 of DWs then competitors must be way off the scale : sweat_smile:i’ll get back to you after 2020 RfQ nominations have been published.
Sitaatti
You assume SEE actually bid for some of your China wins (or China facing).
SEE had to ensure their IP couldn’t be compromised. Hence now that Qualcomm are hard coding it into their system China has now opened up.
This doesn’t really answer anything. “HW agnostic means it wasn’t designed to run consistently, you can’t fix this in customisation.” is a nonsense statement. This is purely a matter of programming. IF consistency is necessary, one can make the software to measure the time every single process takes and then do a necessary syncing to make sure everything runs in a desired manner.
No, it’s not. That would have a completely meaningless effect on performance. The kind of calculations required are nothing compared to what processing and analyzing gaze requires.
Unfortunately, while these systems were sort of adequate for alerting the driver that they might be tired, they are simply not robust enough to determine if a driver is actually watching the road and that is crucial for these more sophisticated partial automation systems. Tesla is a prime example of a company relying only on steering wheel torque to detect driver attention for its AutoPilot system and that inadequacy has been a contributing factor to multiple fatal accidents where drivers used it as a hands-off system which it is not.
Nyt jäi kiinnostamaan että mitä Tesla meinaa tehdä tämän EU-direktiivin kanssa? Integroivatko Smart Eyen Teslaan vai jatkavatko tuolla nykyisellä linjallaan? Teslahan on se yhtiö mitä muut autoyhtiöt yrittävät kopioida
So our foreign visitors are stating that HW agnostism product development doesn’t allow the software to be run in real-time and/or constistically?
And ARM CPU architecture disables the Smart Eyes opportunity of SW oriented business-approach across different platforms…?
What year is it, 1999?
Honestly, this topic is quite technical, please provide some flesh around your statements. At this point they are simply nonsense.
Also, please enlighten us why DMS is dependant of millisecond-level processing cycles. If the driver checks nice piece of ass in the traffic light, the car will emergency brake immediately? Sorry if I sound a bit harsh, but this is something that is concrete enough as an example where we all can understand that millisecond-level processing isn’t necessarily a must have feature.
As the car systems are quite modular today, I’m quite sure that car systems have some sort of Docker and/or Container based technologies in use (or any similar with different name). In other words, as already stated multiple times, HW agnosticism will be key element when securing business for tomorrow in my opinion. Smart Eye has all this covered pretty well.
That kind of delay does not matter in this context. The delay youre talking about is maybe picoseconds. (SEE processor supplier provides that information on their website.)
Generally SW vs HW:
HW optimized solution runs on cheaper HW and with less power. FPGA hardware (which SEE is using) is maybe 10-30x cheaper and more power efficient than standard GPU/ARM/x86 (aka generic chips which run ”hw agnostic sw”). On the other hand designing FPGA chip is rather engineering intensive work, the software cannot really be patched and major releases of software require designing of new chip. (For example if SEE wanted to create interior tracking product, it would probably require engineering of new chip)
Benefits of SW solution are that it is easier to integrate to other software. Also it could be run on shared chip with other things thus reducing HW complexity of the car. That way the SW solution might appear more appealing to car manufacturers. Yes, there are some costs ($5? $50? - dont know) like a need to have bit more powerful chip, but also lots of benefits.
Because its so complicated I think as a non professional who has not worked in the DMS company its really hard to say if HW optimised or HW agnostic solution is better. There are so many variables which are business secrets that determine it. Such as OEM preferences, how heavy you algorithm is etc…
Therefore I am investing to the company which seems to be doing better on the DMS market. Not to some marketing jargon about HW optimisation / agnosticsm.
When reading these stories, I get a sense that Smart has plug’n’play software, which obviously is not true, but it’s customized for each of their customers’ hardware.
What really amazes me is that SM investors (ones that I’ve heard of) do not possess any kind of self-criticism. Personally I try to question my investment cases, well if not daily, at least weekly. I’m ready (and love) to listen to especially people that have different view than mine, as long as they are well grounded.
Time will tell how this battle continues and who gets the biggest pie. But for now, after doing my homework and questioning them over and over again, I decided to opt for a company with hardware agnostic strategy over a company with design win agnostic strategy.
Jos oli JLR:llä ongelmia ja viivästymisiä, niin nyt näyttää olevan uutta tulossa. Katsotaan mielenkiinnolla mitä tulevaisuus tuo seeingille ja smartille
Tästä oli joskus puhetta täällä, (tosin aika vähän muistaakseni) mutta mites täällä sijoittaneet näkee Seyen pitkällä aikavälillä? Puhutaan nyt aikaa 2025v jäälkeen? Tekniikka on todella uniikkia ja varmasti markkinapotentiaalia löytyis muualtakin teollisuudesta? Villeimpiäkin visioita saa heitellä?
Veikkaan että firmat rupiaa käyttämään tuota tekniikkaa mm. Mainonnan vaikuttavuuden mittaamiseen kohdeyleisöllä, kattelin että smartin toimari käynyt kovasti tykkäämässä https://www.affectiva.com/ firman työntekijöiden postauksia, ties vaikka tekisivät jotain yhteistyötä.
Veikkaan että tekniikkaa käytetään jossain google-lasien tyyppisissä jutuissa, joissa katseella toimiva käyttöliittymä.
Veikkaan että tekniikkaa ruvetaan käyttämään lääketieteessä esim. aivoperäisten tautien ennenaikaisessa tunnistamisessa.
Onhan näitä. Smartin sivuilla aika paljon juttua. Esim. Yliopistoissa tehrään tutkimusta.
Olisihan se hienoa jos SE soveltuisi kaikkeen ja kaikkiin, mutta hieman skeptinen olen. Joitain toimijoita nostettu esille kuten Tobii jne. joiden fokus taitaa olla juuri siellä minne SE:n toivotaan laajentuvan.
Smart tekee 2020 liikevaihtoa 7 miltsiä. Ei ole resursseja syleillä kaikkia maailman eri teollisuudenaloja toimitusketjuineen. Pitää muistaa, että tämä kansainvälisesti vielä ihan nano-cap lappu
siihen peilaten pelkkä automotive tulee tuottamaan tasaista, kasvavaa ja ennakoitavaa kassavirtaa korkealla marginaalilla (ohjelmistotalo!) ainakin noin 15 vuodeksi eteenpäin (2035), koska DMS.
Koko DMS-markkinan koko on noin 3.5mrd €, eli 500x puljun liikevaihto, jjosta oletettavasti se 1/3 eli miljardi euroa (muistanko oikein Redin rapsasta ?) aina tulisi uudelleen RfQ-/kilpailtavaksi parin vuoden välein. Redin rapsa pohjaa taas Smartin johdon haastatteluihin ym. jossa toimiala tunnetaan ja nähdään jo pari-kolme vuotta etukäteen toimialan kehitystä ja omaa asemoitumista siinä…
Mä en odota Smartin sijoituskeissiltä paljoa muuta kuin tätä DMS-/OMS-kyvykkyyttä 2020-luvulla. RI-puoli ei edes hirveästi minua kiinnosta, mutta tietysti teknologian olemassaolon ja kehittämisen kannalta sitä on tarvittu nyt ja jatkossa. Itse kyllä tietävät mitä siellä hääräävät.
Ymmärrän hyvin miksi Smart on fokusoinut automotiveen yli 10 vuotta. On ymmärretty tämä kokonaisuus ja että joskus tämä tekniikka tultaisiin vaatimaan autoihin. Nyt se aika on pian käsillä…
Yhtiö on siis kaapaisemassa satoja kertoja itseään suuremmasta markkinasta näillä näkymin sen 40-50%… ei tietysti tarkoita osakkeelle satakertaistumista, mutta johonkin 5-10x voi hyvin tälle laskea pidemmälle aikavälille noin varovaiseksi arvioksi konservatiivisilla kertoimilla
i’ll get back to you after 2020 RfQ nominations have been published - please remember this 
Sorry if I sound a bit harsh, but this is something that is concrete enough as an example where we all can understand that millisecond-level processing isn’t necessarily a must have feature.
