Current Server Time: 03:04 (Central Europe)

#apertus IRC Channel Logs

2013/09/27

Timezone: UTC


09:19
aombk
joined the channel
10:14
aombk
left the channel
10:18
Bertl
morning everyone!
10:35
se6astian
joined the channel
10:36
se6astian
Hello!
10:39
Bertl
hello!
10:46
se6astian
about the FPGA: Could we just leave the space that we don't need in the end as Partial Reconfiguration Area and already build in the interfaces to get image data to and from that area even thought its unused for now?
10:47
Bertl
no, not really
10:48
Bertl
for partial reconfiguration, you need to know beforehand what you want to plug-in and where
10:48
Bertl
think of it like having a physical socket, where you can plugin a chip
10:50
Bertl
but what is the idea of having a plug-in architecture for the FPGA?
10:51
Bertl
I mean, I get it for filters and similar, but you basically know the interface for that beforehand and you usually don't do generalized plugins because of the added complexity
10:52
Bertl
also it gets really complicated if you want to have more than one plugin :)
10:54
Bertl
here, something to read: http://www.wiki.xilinx.com/Zynq+7000+Partial+Reconfiguration+Reference+Design
10:56
se6astian
ok so an app store for the FPGA where you can add new codecs for example is not really possible, the only thing we could do is that you can download one codec at a time only and that contains an entirely new binary FPGA image right?
10:57
Bertl
codec for encoding the image I presume?
10:57
Bertl
(so strictly speaking an encoder)
10:58
se6astian
yes, like a lossless raw encoder
10:59
se6astian
I am thinking out loud if this app store could be of interest to IP providers currently
10:59
se6astian
like IntoPIX
10:59
Bertl
unlikely, for several reasons:
11:00
FergusL
left the channel
11:00
se6astian
they could sell their proprietary jpeg2000 IP core to end users if we allow LGPL embedding of their stuff in our framework
11:00
Bertl
- licensing (how to prevent the IP from being used without license?)
11:01
Bertl
- interfaces (we cannot accomodate all possible encoding IPs)
11:01
Bertl
- optimization and FPGA space (with 4K@150FPS timing is critical)
11:02
se6astian
ad licensing - yes same as with current apps on android/iphone
11:02
Bertl
so you want to make a closed camera now?
11:03
Bertl
which needs to be rooted/jailbroken to run free software?
11:03
se6astian
ad interfaces - they could adapt our code to provide the proper interface for them (only release the changes they made as open source)
11:04
Bertl
well, if they adapt our interfaces, it is basically a complete FPGA reconfiguration, no?
11:04
Bertl
and it won't work for any other codec and definitely wouldn't allow more than one
11:04
se6astian
so you want to make a closed camera now? no - I want to go through some "what if scenarios" to see how we can make everything even more modular
11:06
se6astian
the android scenario is interesting as the system itself is open (though owned by google) but it allows an ecosystem of closed and open source apps to run on it as platfom
11:06
Bertl
no, not really
11:08
se6astian
please elaborate
11:09
Bertl
well, android is open source, yes, it also uses a linux kernel (also open source)
11:09
Bertl
but an android device, by default, is locked down
11:10
Bertl
i.e. you can only install applications certified by google (in some way)
11:11
Bertl
also it makes heavy use of access and permission control, i.e. your app is not able to do certain things, like modifying the kernel or so
11:11
Bertl
(which is important for the security)
11:17
se6astian
but locked devices are the result of manufacturers who implement it, not androids fault...
11:17
Bertl
also note, that on a rooted android device, you can easily install apps you would have to pay for without paying :)
11:18
se6astian
you can actually also installed an *.apk that you would have to pay for on a non rooted device
11:19
se6astian
anyway, I think we are distracting ourselves here :)
11:19
se6astian
the core idea was: can we bring an app store like experience to the camera world?
11:20
Bertl
what I am saying is, no IP provider will attach to an 'open' interface which could easily be used to wrap their IP without any permission from the IP provider, by just copying it from an axiom camera
11:20
se6astian
and curently it looks like we can, but only with official and unofficial full FPGA images basically
11:21
se6astian
like you can get custom ROMs for your rooted android device
11:21
Bertl
yes I think we could do that but I'm also sure that we do not want to do that because it would be a lot of work for little gain
11:21
Bertl
(the app store experience)
11:22
Bertl
I doubt that IP providers will allow unsigned FPGA images :)
11:22
se6astian
I see
11:23
Bertl
again, nothing would stop the end-users from sharing any IP
11:23
Bertl
(be it partial bitstream or complete FPGA image)
11:25
Bertl
but yesterday, I thought about the axiom and where you could actually make good use of such a device besides the obvious 'making movies'
11:26
Bertl
and I realized, that it would be the perfect device for machine vision and control
11:27
Bertl
(mainly because of the open platform and the ability to tinker with hard and software)
11:27
Bertl
(assuming that 'we' do not decide to go the proprietary path :)
11:29
se6astian
yes, Konstantin also had some ideas about scientific/inspection use cases right at his workplace (swiss watch manufacturing) :)
11:30
Bertl
so it might make perfect sense to have two or even more software branches for the same hardware, i.e. one for movie makers, the other one for scientists, the third one for solution providers ...
11:31
se6astian
if we do get customers in that market segments we can also afford maintaining such branches yes
11:32
se6astian
but I know nothing about these use cases tbh
11:32
Bertl
which again brings be to the conclusion that if we manage to make a reasonably good product, it doesn't have to be cheap that cheap and more importantly, we can break it down into small pieces and fund each of them separately
11:32
Bertl
s/be/me/
11:32
Bertl
s/cheap that/that/
11:33
se6astian
but there are already some industry cameras even with the same image sensor
11:33
se6astian
like http://www.vieworks.com/homepage/mv/product/product_view.php?sn_series=6
11:34
Bertl
do they allow to run your IP on them?
11:34
Bertl
do they allow you to run your IP on them?
11:35
Bertl
are they cost efficient?
11:35
Bertl
btw, interresting that they do 180fps on the CMV12k :)
11:35
se6astian
yes, old sensor specs :)
11:36
se6astian
they all come with an SDK but I have no idea what that allows you to change
11:36
Bertl
well, let's put the it other way round:
11:36
se6astian
Arent most automation/industrial applications centered around interpreting/processing the images on a host pc?
11:36
Bertl
let's say, you are a solution provider, and oscar is the customer
11:37
Bertl
you pick the VC-12MX-M/C 180 and want to 'make' a movie camera with it
11:37
Bertl
what will be required? what will be the costs?
11:40
se6astian
a "solution provider" is like a product developer but for a single customer only right?
11:40
Bertl
basically, yes
11:41
Bertl
somebody who does all the work required to get a custom solution
11:41
se6astian
I see, yes then Axiom is very attractinve
11:41
Bertl
now how many companies use off the shelve products for machine vision?
11:41
se6astian
also because I dont need to publish any code changes I make because I only use the result "in house"
11:41
se6astian
like what google did with the Elphel cameras
11:43
Bertl
for example
11:44
Bertl
and IMHO that will also be the most interesting area for revenue (i.e. providing solutions based on the axiom)
11:45
se6astian
because custom solutions are in general very expensive and if the camera is now 5000 or 20.000 will not make much of a difference for the end sum anyway right
11:46
Bertl
precisely, and it doesn't matter if we actually build the camera or not, as long as we are 'the axiom guys'
11:50
se6astian
how do you mean "it doesnt matter if we build the camera or not"?
11:51
Bertl
well, I think we will not be able to make money from 'building' the camera (i.e. from the hardware itself)
11:51
Bertl
I do not mean that we should not build it at all :)
11:52
Bertl
but I think axiom/apertus would be fine if after the second axiom prototype somebody steps up to build the entire thing for us (not even paying any royalties)
11:53
se6astian
but those solution providers would also just buy it from that someone who builds the Axiom in the end, how will we get any money from it?
11:54
Bertl
because they will either sub-contract us (for development) or because we actually _will_be_ those solution providers :)
11:55
Bertl
despite the fact that the code and hardware design is public, a solution provider does not want to dig into details on every platform (s)he uses
11:57
se6astian
I see
11:59
se6astian
Elphel has received several of such requests: "can you make a custom camera based on your elphel camera for my application"
11:59
Bertl
and did they do it?
12:00
se6astian
but always turned them down because they want to create a general-purpose-camera only and leave the customizations to others
12:00
se6astian
since everything is open and available anyway
12:00
se6astian
they are looking for value-adding-applications
12:00
Bertl
hehe, now I understand ...
12:01
se6astian
like mediworks, a company in china is building microscopes with built in elphel cameras
12:01
se6astian
not sure if they were successful, but the idea is to have no work but grow sales :)
12:05
Bertl
so, IMHO it is important to focus on keeping it simple but extensible and more importantly adaptable for all kind of crazy uses
12:06
se6astian
yes
12:06
Bertl
and personally I wouldn't bother with anything security or proprietary related because that just adds a lot of workload with little gain
12:07
se6astian
which did not really work out for Elphel because the camera is rather complex and take a lot of time to get into, beside google there is no big customers who did it
12:07
se6astian
ok, time for lunch
12:07
se6astian
bbs
12:07
se6astian
great insights, btw :)
12:40
se6astian
ok and now I need to do some lecture hall installations
12:57
Bertl
do not install too many halls :)
12:59
se6astian
left the channel
16:47
se6astian
joined the channel
16:47
se6astian
time to go home :)
16:47
se6astian
left the channel
17:22
FergusL
joined the channel
17:32
Bertl
welcome FergusL!
17:43
se6astian
joined the channel
17:44
Bertl
wb se6astian!
17:46
se6astian
thanks, back home now :)
17:46
se6astian
I am summarizing the appstore discussion we had on the iwiki
17:47
Bertl
I am working on axiom alpha :)
17:48
se6astian
good, cary on :)
17:48
Bertl
thanks!
18:04
se6astian
https://iwiki.apertus.org/index.php?title=AppStore
18:06
se6astian
my reason for the "app store" is that its a very good way to illustrate what open source can do for people - it illustrates the concept of extendability and ease of use using something everybody knows/experienced already
18:07
Bertl
seems to require access credentials
18:08
se6astian
aka opensource = you get new stuff easily
21:24
dmj_nova
left the channel