#apertus IRC Channel Logs

off to bed now ... have a good one everyone!

morning folks!

hey

So i can not respond for safety reasons :)

Hello everyone!

evening everyone!

se6astian|away is ... well away :)

note that the first evaluation is upcoming, so it is a good time to give an overview what has been done and why not :)

or just previous week?

okay, thanks!

from my side there isn't much to report, I was unexpectedly busy with life end of last week and the weekend

Reports are to be submitted? Where?

here

Okay.

sure, go ahead

Okay the basic model of the sensor is done

And from now on new features are constantly added to the sequencer itself

Every*

More recent updates---

Added mask on non readable register

yes, tested

i thought dnc registers are writable but should not be changed

This month main focus will be on the pattern generator, after discussing that with BertL i’ll upload detailed architecture of the generator and start coding it.

every module has its test bench, except serializer

DONE

thanks for the report, as final words, how does that compare to the original timeline?

AHEAD

that's always good to hear :)

hello everyone, good evening!

i.e

(2) Sobel convolution filter

after that I would be able to claim the constant delay in the pipeline because of my overlay module.

This leaves me enough time to implement subtle changes to be suggested in the future as well as space for FPGA implementation.

thats all from my side.

any suggestions or comments on the architecture, algorithm as well as the coding style/technique is welcomed.

nmdis199-: the channel is yours now!

Here's description of my work and what I achieved / did not

I started reading and understanding about architecture of microzed and trying to understand cache optimization during which I also worked on figuring out about “triggering image capture”.So, till 25th of May (which was my exam week plus the works I mentioned I did above) this was my work.

https://docs.google.com/document/d/1vCjhKgLnSdHelyPBqxCVb9fpOz9EoaNPyYiR_lG_l8g/edit?usp=sharing

I originally had to improve cmv_hist3 and to introducing image crop, before this I had to know how the codes are working on beta for improvement.

There were 7 versions of the code, which you can find in .../opt/ITI folder in beta , also two tests are uploaded on my github repo

https://github.com/nmdis1999/ReadFrameBuffer

here -> https://docs.google.com/document/d/1Za5jm8m_NJNbg_K4-mtbtJjfW7qSxrKeIxp_8Ron2Aw/edit?usp=sharing

I wanted to work on the crop tool (note I already implemented on in the histogram tool) so, that was left. Other than that, I am now able to understand how to work on beta and result of our codes on different versions on beta.

Thanks everyone, that’s all :)

compared to the original timeline, how is your work going?

except of "crop" tool in cmv_hist3 I did others, including the "histogram" tool which was an addition. So, pretty much I am going with timeline

7pm I mean, under 30 minutes

TofuLynx: okay, we will be around and listen to your report

you're welcome!

sure, go ahead, channel is yours now

Link to progress report. https://docs.google.com/document/d/1oV-F7XNQPQVUzSAbNNtzppTzIOryoggFRxhOTabgTnY/edit?usp=sharing

receiver github repo link. https://github.com/Yayi98/dec_prng_receiver

okay thanks for the report! appreciated!

yes, please go ahead!

1. understanding if there is a general standard LUT specification

2. I had completed the task https://lab.apertus.org/T1018 finally

regarding GSoC,

Analysis phase ends: report findings and suitability of each container format. Report if custom containers are feasible.

eval 2 says: ​Provide a functional and working system that can encode/decode the decided video container format.

but CDNG are still to come.

we are definitely over eval 1 and would touch eval 2 work soon

@Tofulynx, thanks

for the link: this was the vid: https://www.youtube.com/watch?v=Rpq_ItPHFx8 :)

Same feedback from my side, like the editing.

that was long done... g3gg0-mobile

but I toasted out my computer becuase it was not able to record on nvidia optimus

had to reinstall everything.. and lost 300Gigs of data (not important though) ouch

(in the meantime)

much better than last year, many active and eager to learn people around

First time for me, despite the time zone difference and mail/chat only, i can see good progress

Students will need this experience later, and employers love to see it

20 years old and 15 years experience in the industry ;)

that's what HR await

how is that possible?

Can I report my progress now?

xD

It was a troubled experience, with a lot of mistakes from my part in the code xD

However, I fixed the mistakes and the image got better

I proposed to the creator of the algorithm to implement a sort of edge sensing algorithm

Overall, the result looks more sharp, but at the same time, it's fuzzy

I suspect that it will give a sort of grainy look to movies and moving images

so we can test it out

yeah, hope so

I commited to implement moving pictures if I finish my proposal sooner than expected

so do you like SHOODAK?

in my opinion: it's hard to tell without motion, but edges seem to be aproblem, GEDI is much clearer

maybe we can do a combination of GEDI and SHOODAK

BAndiT1983: the combination is pratically done

In that regard

yeah

but, if you look at SHOODAK code

haven't looked yet, will do it this week

and it will be, in my opinion, a better result than GEDI

yes, please

@BAndiT1983, I definitely agree, I have learned more in last 2 months with the help of mentors than I did on my own, besides I have learned to organize my work in good way. Also, as I already mentioned I indeed admire apertus's way to have regular meeting and the way everyone (including students participating here ) is active.

I second you, nmdis199-

okay, so next meeting next monday, same time, same place!

g3gg0-mobile: anything for me?

meeting for today (a longer one) is now concluded ... feel free to chat on

GEDI is still a bit more crisp, or am i wrong?

I said that :p

http://graphicon.ru/oldgr/en/publications/text/prog2004lk.pdf

?

i have also to learn about the algorithms, as i have to check where we can gain performance and also if the code has to be adjusted to be maintainable or similar

I already confirmed that what I used is not that. SHOODAK is uncommon because the new pixels are in the center of the 2x2 block, and as such it's different from those which just fill the missing pixels.

so NEDI is different to GEDI?

openmp is just the first step, we target parallel processing through opencl

NEDI os different to GEDI, it's an improvement over GEDI. It uses, for example, bigger neighborhood size

image*

what do you mean?

Huhm I am asking if the displayed image is the result of a one-time demosaic or is always demosaicing the original image continuously

Bertl there´?

cool

from the datasheet, the sensel's initial value is Vhigh, what is it set to?

most likely the pixel voltage supplied to the sensor

(minus whatever the mosfets take)

okay thanks

ok. we could try to make it always rendering a new image to see what effect SHOODAK brings to moving images

we need real sequence, otherwise you will have no effect

TofuLynx: what's the plan currently? openmp?

very good, don't forget to test against release mode, i can tell you how to switch to it, if you have trouble

do you need further improvements or extensions in the UI?

it'S 8bit, as lodepng cannot output 16bit RGB somehow

and not image center based

yeah, I used PNG dump to send you the comparison and samples to Stephan

UI has small bug, as the theme turns disabled button black, but it's minor stuff for later

good idea :)