#apertus IRC Channel Logs

Sorry wrong link

And https://drive.google.com/open?id=13Nm98KCBIYZPXKWEQmxrlrY60WjrHYtG

excellent! will check tomorrow ... off to bed now.

Bertl_oO: Can we use the single IOs for signalling the Zynq from the MachXO2 that the link has been trained?

rbeiten unter anderem mit dem OFAI in Wien zusammen. Jetzt suchen wir "Co-player", die Lust haben, mit uns zusammen, die Software, Holly (die KI) und auch die Machine Vision-Skills zu entwickeln. Dafür haben wir unsere ganz eigene Stellenausschreibung verfasst. Die findet ihr hier: https://www.playsys.at/join_us/. Würde mich freuen, wenn es die ein

morning folks!

off for now ... bbl

Greeting BAndiT1983 ,

supragya_, will also join , may be

This videoclip class would contain parameters like debayering algo to be used, quality used for downscaling(if required) also methods for changing them, either by user or other modules(otherwise they may contain default values). once we have knowledge of processes(like downscale by 1/3) to be used on frames, we can easily allocate memory according to that( use InitAllocator() which can easily allocate size in mutliple of framesize).

The class should have method which can be used to read bulk of files (each acting as frames) from a directory. I was thinking to use a large memory block agian and agian for each file and then processsing it. Or may be anything better we can come up.

info's related to processing methods to be used on frames

does video clip really need it?

isn't the whole process a pipeline?

It will be used for intraction between OCcore and frameserver

video clip class

as i said, we have a pipeline, so why should the infos be placed in video clip class, which do not relate to it

what is the process, when user changes downscaling or any other parameter from frame server UI?

Good evening sraj

it doesn't matter which one, as they derive from same class

Gobi? Where's that? Anyways... Meeting meeting

https://en.wikipedia.org/wiki/Gobi_Desert

India - Pakistan border

there are several there, according to wiki

My dad's posted here

ah, i remember, military

Hmm... What I was trying to tell Dev was just that

Continue, dev

let me through in another term: event bus ;)

It's better if you explain your whole thought on this matter Andrej

ok, ok, thought that the code would be known already

i've implemented event bus a long time ago, to be able to aps messages between different modules, you can inspect OCBacckup code for direct examples

It needs it during memalloc and then we need to freeze the settings in Oc

it will help us to allocate memory

required memory*

video clip could contain resolution and frame list, frames contain integers which are indexes to buckets in static allocator

i mean the indexes to buckets set, e.g. RGB

https://dzone.com/articles/design-patterns-event-bus

Dev, please save these links for reference

logs are also there

this allows to avoid to call it from presenter and also it could be used to trigger multiple things in the system, sadly i don't remember it that well at the moment

In easy way think of it like a event triggering functions where callbacks are registered

*event triggered function

nice examples in the docs there -> https://github.com/gelldur/EventBus

Hello se6astian

Or one has to look into testing that as well

but we need it, as otherwise it would result in some spaghetti code

my qquestion is still not answered, how do we allocate buckets of required frame size and not more?

https://github.com/kakashi-Of-Saringan/opencine/blob/dev/Source/OCcore/Memory/StaticAllocator.cpp

what is the quality parameter?

There's 1/1 as well :P

but that's not our result, what does it result in, when we set the parameter for downscaler?

like , if I set 1/2 as a parameter , we only need h/2 * w/2 as framesize . The buffer of static allocator will be allocate size in mulyiple of k * framesize

15MB maybe too less... Just saying #hyperparametertuning

i don't care about memory sizes at the moment

if we set 1/2, 1/3 etc., then our frame size shrinks in terms of width and height (let's leave bits per channel out for now), which is important to us, as we have to tell allocator to calculate bucket offsets according to this sizes

so we want downscaler to tell the allocator about this parameters , yes

Dev

we have loader -> processor -> allocator and so on

as we know that loader is loading, then allocator has to allocate and frame processor to process, but we also know that frame processor could reduce the data amount, by using only every second pixel

task is to implement proper point, so the allocator can tell that sizes are different, no events or other voodoo, as we are on the same layer for now

While downscaler can downscale 200*200 to 100*100 or let 100*100 remain 100*100 what matters is whether the final frameserver endpoint knows about the downscaler param...

And yes, it happens just the first frame

Well here goes the answer

Two options are there

Either alloc something and manage pointers (bad way)

Both had pros and cons

Yes, I am here

what should this data structure contain

so it's fine to allocate 1 chunk, like 512MB or so, then dissect it into buckets, depending on loaded frame size, if video changes just calculate new buckets, nothing difficult there at basic level

#mappings

in our case, if we consider uin16_t data, we need 3 buckets per frame

but we can also use the allocator for uin8_t, which doesn't matter much

512MB is fine

Especially for the file sizes we deal with

But that's extension for later date

dev__, are you there?

any comments?

i thought that we discussed it before and also your gdocs contains similar graphics

so it's fine to allocate 1 chunk, like 512MB or so, then dissect it into buckets, depending on loaded frame size, if video changes just calculate new buckets : All this is implemented in statica allocators

Bertl : ping... There's gotkraken being used :P

So are u helding out videoclip class

:D

dev__, what do you mean?

do it as soon as possible please, as i want to get more progress this month

trello is the fastest way, as i can see it on my smartphone at work

update u soon .

DGRAM and STREAM don't seem to work rn.

STREAM should work, but maybe you need to adjust both sides

e.g. check if data is incoming and read as long as the packet is not received fully

maybe you can check if it is possible to open 2 sockets, one for dgram another for stream, as small packets are better suited for dgram

here are also some infos on both -> http://www.linuxhowtos.org/C_C++/socket.htm

Had a look through all of these, but first the question remains that do we need this in first place .

If we were to send 4096*4 float values for lut_conf ,this can be easily sent over as a single packet.

you can of course raise the size up, but please ensure that it's not slowing down

I will clean the code and push it ,maybe you can check if it works according to the requirement