Current Server Time: 18:36 (Central Europe)

#apertus IRC Channel Logs

2019/03/28

Timezone: UTC


23:00
Bertl
off to bed now ... have a good one everyone!
23:00
Bertl
changed nick to: Bertl_zZ
23:04
futarisIRCcloud
joined the channel
00:24
Spirit532
left the channel
00:25
Spirit532
joined the channel
01:00
intrac
left the channel
01:01
intrac
joined the channel
01:06
intrac
left the channel
01:06
intrac
joined the channel
02:16
shivamgoyal
joined the channel
02:50
shivamgoyal
left the channel
03:51
Spirit532
left the channel
03:52
Spirit532
joined the channel
05:33
BAndiT1983|away
changed nick to: BAndiT1983
05:39
Bertl_zZ
changed nick to: Bertl
05:39
Bertl
morning folks!
05:40
siddhantsahu
morning!
06:01
siddhantsahu
Bertl: Kindly you guide on the implementation details for T1142
06:03
siddhantsahu
guide me*
06:03
Bertl
ah yes, thanks for reminding us ...
06:03
siddhantsahu
I wish to write a proposal for GSoc 2019 for this task as I am interested in this topic.
06:04
siddhantsahu
I have completed the T872 C++ challenge
06:05
Bertl
so, I can't speak for OpenCine, BAndiT1983 is the one you want to contact there, but what are your ideas for the AXIOM Beta side?
06:05
siddhantsahu
Working on the BMP class for T872
06:15
siddhantsahu
AXIOM Beta users can have the option to select an automatic white balance algorithm so that the scene's illuminant color can be predicted and the color cast can be removed by doing some linear color transformation(von kries hyposthesis). Since AXIOM Beta records RAW files, color temperature for footage can be easily modified in post, but for preview(in monitors) we need to specify a color space along with a white point
06:15
siddhantsahu
according to which the RAW files colors are mapped. Users can the option to manually dial in the color temperature for the scene or they can adjust it in post (OpenCine) using any of the presets (Cloudy, Daylight, Incandescent..etc) or automatic one, which I wish to implement.
06:16
Bertl
okay, so how do you plan to do the white balance on the preview?
06:17
siddhantsahu
As far as I have read, LUTs (look up table) are usually used for color and color space transformation.
06:18
BAndiT1983
changed nick to: BAndiT1983|away
06:18
Bertl
okay? and how do you calculate the settings in realtime?
06:21
siddhantsahu
https://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Cheng_Effective_Learning-Based_Illuminant_2015_CVPR_paper.pdf
06:22
siddhantsahu
Here is the state of the art automatic white balancing algorithm which is learning based model (training required), once trained the inference time (0.3 secs) is at par with other classical methods.
06:23
Bertl
0.3 secs on what hardware?
06:23
Bertl
also what resources and input is required for this to work?
06:25
siddhantsahu
Yes, the runtime is calculated on a Intel Xeon 3.5 GHz processor. Certainly this won't be case for AXIOM Beta's processor.
06:26
Bertl
indeed, you are probably a factor of 10-100 off there
06:26
siddhantsahu
But wouldn't this implementation of Automatic Balance should work fine for desktop OpenCine user with faster processor?
06:27
Bertl
certainly, so you want to drop the Beta part in your task?
06:28
siddhantsahu
As far as I know the paper link which I provided is the fastest learning based algorithm.
06:28
siddhantsahu
Yes will do that
06:28
Bertl
okay, then please do so, also add some explanations why
06:29
Bertl
and have a chat with BAndiT1983|away once he returns regarding OpenCine integration
06:29
siddhantsahu
Some of the classical AWB which might work in the case of Beta are "Gray World" and "White Patch".
06:30
siddhantsahu
Yes will ask BAndiT1983 when he is around
06:33
se6astian|away
changed nick to: se6astian
06:34
se6astian
good day
06:40
siddhantsahu
Bertl: changes the task (T1142) with explanation.
06:56
se6astian
changed nick to: se6astian|away
06:57
se6astian|away
changed nick to: se6astian
08:49
Topic
apertus° - open source cinema | www.apertus.org | join the apertus° Lab: http://lab.apertus.org/ | IRC Logs available at: http://irc.apertus.org
08:49
se6astian
has set the topic
08:50
rahul-
joined the channel
08:50
nmdis1999
joined the channel
08:51
se6astian|away
joined the channel
08:51
se6astian|away
changed nick to: se6astian
08:51
ArunM
joined the channel
08:51
RexOrCine|away
joined the channel
08:53
Bertl
joined the channel
08:56
BAndiT1983|away
joined the channel
08:56
BAndiT1983|away
changed nick to: BAndiT1983
08:56
Bertl_xX
left the channel
08:58
philippej|away
joined the channel
08:58
philippej|away
changed nick to: philippej
09:31
Ashu
joined the channel
09:39
Ashu
left the channel
10:10
Y_G
joined the channel
10:11
comradekingu
joined the channel
10:14
Y_G
Hi ,regarding the task "Add further camera/image sensor parameters, besides currently implemented image sensor plain register configuration, digital and analog gain, like white balance, color matrix for output etc." from T1121 ,my understanding is that "mat4_conf.sh" is currently being used for white balancing and we want to improve this
10:15
se6astian
well mat4 just sets per color factors
10:15
se6astian
but doesnt do the actual whitebalancing
10:17
Y_G
Thanks,are there any directions how we should go about doing actual white balancing
10:17
se6astian
can you explain to me what the whitebalancing process is?
10:19
Y_G
removing unrealistic color casts?
10:21
se6astian
how?
10:21
Y_G
I have gone through this only regarding white balancing "https://www.cambridgeincolour.com/tutorials/white-balance.htm"
10:23
Y_G
Have'nt dived into the how part as of now,this is where I was looking for help
10:24
Bertl
any work done on the qualification task?
10:25
Y_G
@Bertl Yes,I have to clean the code and work on the "nonlinear curve for the 12 to 8 bit conversion"
10:26
Bertl
great!
10:27
Y_G
@se6astian I'll get back to you on the how part in some time ,sorry for asking questions without diving into it :)
10:31
Ashu
joined the channel
10:34
Y_G
left the channel
10:53
Ashu
left the channel
10:54
Ashu
joined the channel
11:32
siddhantsahu
BAndiT1983: Hi, can you pl help me with the implementation for T1142
11:33
siddhantsahu
Check today's log with Bertl
11:35
Ashu
left the channel
12:11
Bertl
off for now ... bbl
12:11
Bertl
changed nick to: Bertl_oO
13:12
futarisIRCcloud
left the channel
13:54
RexOrCine|away
changed nick to: RexOrCine
14:10
aombk3
joined the channel
14:13
aombk2
left the channel
15:21
sebix
left the channel
16:15
niemand
joined the channel
16:15
niemand
left the channel
16:15
niemand
joined the channel
17:01
BAndiT1983
hi siddhantsahu, do you need something specific?
17:03
siddhantsahu
Hi BandiT1983
17:13
siddhantsahu
For the task T1142 I want to implement this https://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Cheng_Effective_Learning-Based_Illuminant_2015_CVPR_paper.pdf algorithm for OpenCine and compare it against the classical methods.
17:15
siddhantsahu
classical algorithms for automatic white balancing (Gray World, White Patch..etc)
17:16
siddhantsahu
This is another state of the art learning based AWB http://openaccess.thecvf.com/content_cvpr_2017/papers/Hu_FC4_Fully_Convolutional_CVPR_2017_paper.pdf
17:18
siddhantsahu
To implement these algorithms I can build the neural architecture in some framework (Tensoflow, PyTorch) and train the model if the pretrained weights for the model are not available. After training I will get the trained weights for each layer and the network configuration file.
17:20
siddhantsahu
After that what should do?
17:21
siddhantsahu
I have read about how to use trained models for inference in OpenCV dnn module.
17:22
siddhantsahu
https://www.learnopencv.com/deep-learning-based-object-detection-using-yolov3-with-opencv-python-c/
17:22
BAndiT1983
don't know how i can help you there
17:25
siddhantsahu
I mean we can use use any trained models weights and config file and do the inference on our image with the help of the OpenCV dnn module.
17:26
se6astian
changed nick to: se6astian|away
17:26
siddhantsahu
To implement it for OpenCine, should we use OpenCV dnn module dependency or we need to build it from scratch?
17:26
siddhantsahu
Please refer this https://www.learnopencv.com/cpu-performance-comparison-of-opencv-and-other-deep-learning-frameworks/
17:27
siddhantsahu
Inference on Intel CPUs with OpenCV dnn module is very optimized.
17:28
BAndiT1983
this is something you have to tell me, as i have not much knowledge about this stuff and need more info, than just some abbreviations
17:29
siddhantsahu
Which part, please ask
17:30
BAndiT1983
everything, jow would you bind it to OC, at which stage of pipeline should it happen, what is involved?
17:30
BAndiT1983
*how
17:31
siddhantsahu
I am still reading the OC code base and yet to understand it completely
17:32
siddhantsahu
Would be great if I can refer to any dataflow diagram documentation for OC.
17:32
BAndiT1983
yes, would be great, if someone would have time to sit down and create one
17:33
siddhantsahu
Basically these AWB (automatic white balancing) algorithms can be applied on imported RAW footage from the Beta in OC
17:34
siddhantsahu
I will make one once I go through it completely.
17:36
siddhantsahu
Please refer this link
17:36
siddhantsahu
https://www.google.com/search?q=lightroom+white+balance&rlz=1C1CHZL_enIN748IN748&source=lnms&tbm=isch&sa=X&ved=0ahUKEwj0ioWevaXhAhVdi3AKHVPvAzMQ_AUIDygC&biw=1920&bih=980#imgrc=45U2YGiTntQisM:
17:37
siddhantsahu
You will see a "auto" option
17:39
siddhantsahu
along with some presets (Daylight, Cloudy, Shade..etc)
17:39
siddhantsahu
and a "custom" option as well.
18:28
aSobhy
joined the channel
18:44
se6astian|away
changed nick to: se6astian
19:05
BAndiT1983
changed nick to: BAndiT1983|away
19:58
BAndiT1983|away
changed nick to: BAndiT1983
20:35
se6astian
changed nick to: se6astian|away
20:46
BAndiT1983
changed nick to: BAndiT1983|away
20:47
niemand
left the channel
22:15
futarisIRCcloud
joined the channel
22:28
Bertl_oO
off to bed now ... have a good one everyone!
22:28
Bertl_oO
changed nick to: Bertl_zZ