3K/UHD 5D2 Raw development and Other Digic IV Cams

Started by reddeercity, April 06, 2017, 12:22:27 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

DeafEyeJedi

Fantastic work and thanks for sharing @2blackbar!
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

reddeercity

Working on 50D x1.67 stretch (720p Hi frame rate Mode)
Found the adtg reg to change the vertical to the 1.67 pixel (squeezed)
adtg[12]100c 0x2 ->0x4
same as the 5D2 , I also reduced rolling shutter .
Head times are a little touchy but was able to get 50p for a brief time then lock up ,
I should be able to get a stable 48p at some point but the math says I should be able to get 1568*576*96fps=86 Mpixel/s(mega pixels per second)
So 96fps should be possible I think if all thing are equal  ::) since the sensor speed is 115Mp/s
I short test with 1.67 stretch but with only 27 fps ( to be sure it work before I tried to increase frame rate to 50p+)

50D_1.67_Stretch_1568x576_M14-0117.MLV 118MB file (4 second of raw)


1568x576_1.67_stretchM14-0117_frame_103_1.png


1568x960_1.67_stretch_M14-0117_frame_103.png

2 , 4second clips one squeezed & the other still squeezed
1568x576_27fps_1.67_squeezed_M14-0117.mp4  1568x960_27fps_1.67_un-squeezed_M14-0117.mp4]

reddeercity

Updated the 3k preset (3008x1080) with the correct horizontal centering as per this post  here
magiclantern-crop_rec-3k_Updated_Center_4.20pm-5D2-eXperimental.2019Nov14.5D2212.zip
Edit: updated the first post on page one with updated 3k corrected centering

pyffi

hello! in days i saw the great news about 4k and FHD 48fps on 5dmk2
download latest build in this conversation
and modules doesnt work(
- format and install on clear CF NO WORKING
- replace modules NO WORKING
and i see this message


can you help me? and what i doing wrong?

5dmk2+CF Transcend 32gb 1000x

dariSSight

Thanks Again,  Can we use 3x3 1856x1248 with magiclantern-crop_rec-3k_Updated_Center_4.20pm-5D2-eXperimental.2019Nov14.5D2212.zip, or do we only use it for crop and also crop rec is more useful with crop lens or 24mm, right. Its so much redundancy on the front page, so if you can so I won't need to ever ask these question again, please answer these questions: 1. which firmware & tutorial is for the most stable 48fps, 2. which firmware & tutorial is for the most stable 3x3 FHD 1856x1248 3. which firmware & tutorial is for the most stable Crop-Rec 3008x1080 4. which firmware & tutorial is for the most stable 3x1 and 5. which firmware & tutorial is for the most stable 1x3.

I know these questions might irratates you but PLEASE MAKE THIS CLEAR for user so we can better utilize as instructed, THANKS for any info you can clear up.
Canon 5D Mark II

reddeercity

Working on pixel binding 50D trying to remove & or fill that black bar with data, but found something strange today .
I can some how get 1:1 pixel in 3x3 mode  :o
I set cmos[4] (horizontal pinning) on 50D cmos[2] on 5d2 & 5d3 to 3xcrop_mode
cmos[4]0x40c(default3x3) -->0x434 (3xcrop_mode 1:1)
Then I set the c0f06084 to 3xcrop_mode (default)
c0f06084 0x10045(default3x3) --> 0x1004a(3xcrop_mode default)
Finally I set adtg[12]1000 to default 3xcrop_mode
adtg[12]1000 0x6(default3x3) --> 0x05 (default 3xcrop_mode)
on other D4/5 cams you need to set adtg[1]100c (d4) or adtg[1]800c (d5) line skipping mode
but on the 50D it looks like you only need adtg[12]1000 to change line skipping to non line skipping
After setting raw image size to 4096x3166 @ 5.4 fps (locks up/freezes @ 8fps) ok at 7.5fps
I have a full height with near full width (I didn't want to go full width yet) and without any aliasing  8)
So it seem the 50D is a different kind of beast ! thou it similar to 5d2 & other D4 cams it has some
unique challenges .




M18-0012_00001_4096x3166.png
M18-0012_4096x3166_5.4fps_00001.dng

100% zoom in mlvProducer viewer , no  aliasing


200% zoom in mlvProducer viewer and again no aliasing


Edit: here the MLV I was testing 4096x3166 @ 5.4fps  M18-0012_4096x3166_5.4fps.MLV
The upper portion before the black bar is equal to the liveview in 3x3 (1067 lines or vertical pixels) and in 3xcrop_mode the upper portion is again = to Liveview which is 1089 .
it seem this has something to do with the 50D video canon hack I think , but I can't proof it the silent simple dng's have the same black bar but the Full Res silent Dng's are fine , no black bar .

SKecorius

Dear reddeercity (Dave?),
fantastic work! Thanks a lot for all your time and efforts. I have recently purchased well preserved 5D2, and now discover the world of Magic Lantern.
I would like to ask, if "magiclantern-crop_rec-3k_Updated_Center_4.20pm-5D2-eXperimental.2019Nov14.5D2212.zip" is the last working build that would allow me to use (assuming I have fastest lexar cf) the following:

1. continuous 1044 raw at 23.9 fps
2. 12-bit and 10-bit full 1080 raw at 23.9 fps
3. higher resolution (3K, 2.5K) using cropping, at 12-bit, 10-bit 23.9 fps, that would reduce moire and aliasing
4. dual iso video at 23.9 fps (I saw some videos online that say dual iso can only be used for static shots (due to ghosting)).

I am eager to test the mentioned functions and perhaps to conclude my experience in a short report here in forum.

I would also like to ask:

a) what is the biggest capacity CF that will work on 5DmII with ML?
b) how far is your eXperimental 2019Nov14 build from becoming a nightly build?
c) do you have any other plans that are one the pipeline for 5DmII (e.g. improvements on the existing features, rather than adding new ones)?
d) is there a list of known issues that comes with your latest 2019Nov14 build?

Looking forward hearing from you.

Sincerely,
Simonas

reddeercity

Looking in to that black bar on 50d , and there is data there

300_percent_zoom_in.png

M19-0024_00001_black_bar_full-res_4448x246.png

so I thinking now that there's a corrupted & or incorrect address for that raw stream part of the sensor
If I'm not mistaken , there 3 different stream address , from the "consts.h"
consts.h#lines-44
#define YUV422_LV_BUFFER_1 0x41B00000
#define YUV422_LV_BUFFER_2 0x5C000000
#define YUV422_LV_BUFFER_3 0x5F600000

maybe buffer_2 ? is the problem
then I see this
consts.h#lines-54
// not 100% sure, copied from 550D/5D2/500D
#define REG_EDMAC_WRITE_LV_ADDR 0xc0f26208 // SDRAM address of LV buffer (aka VRAM)
#define REG_EDMAC_WRITE_HD_ADDR 0xc0f04008 // SDRAM address of HD buffer (aka YUV)

the LV addr is from the 5d2 etc... so how can I verify these address to be 100% correct
I never look for addresses before , so not too sure how fine them .


histor

No, that looks like actual data from the sensor. I get such bias frames adjusting analog iso register (between the main steps) (and I successfully use them as bais frames). Maybe you managed to set different iso for this part )

reddeercity

Ok I think I finally found on to fix the black bar (kind of) or I should say I found how to change the lower half
to black out a portion or the lower screen just like the upper half , see the raw image dumps below
So it looks like the ADTG1[1179]N
Controls the lower portion of the screen
   
ADTG1[1179]N 0x53A ->0x1A5 , black the whole lower portion                      ADTG1[1179]N 0x53A ->0x5BA       
RAW-014.png                                                                                                         RAW-015.png
   
ADTG1[1179]N 0x53A ->0x845                                                                   ADTG1[1179]N 0x53A ->0x9BA
RAW-016.png                                                                                                            RAW-017.png
   
ADTG1[1179]N 0x53A ->0x5A4                                                                   ADTG1[1179]N 0x53A ->0x45B
RAW-018.png                                                                                                        RAW-019.png

So I just need to find the upper portion ADTG1 Reg now .

Edit: Also here a very short MlV 450MB 4k-Dual_ISO_50D_M19-2349.MLV -- found out that cmos[0] enables it ,
not to sure what recovery ISO is the base was 800ISO , it by luck I found it I was looking for reg's for the black bar. 

reddeercity

Getting somewhere now on the 50D (black bar problem) seems I need to use 2 different "ADTG" Regs
adtg1[1172]0x665 -->0x465
adtg1[1178]0x665 -->0x465





4134x2593_RAW-022.png
So now I can get vertically 1950 out of the 2593 so 643 lines at the bottom ,
much better now at least I can get now 3100x1300 in 5xZoom


4134x2593_RAW-022_croped_to_4108x1950.png
I just cropped off the bottom to view it without the black bottom .
I still need to find the other reg's to let me have full view height .

Luther

Nice progress here @reddeercity ! I still need to try the last build, but the first one work well.

DeafEyeJedi

I seem to have better luck with the latest build as opposed to the older one for the 50D. Been a few weeks since I've played.
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

Macke1

New Update: November 14/2019
3008x1080 @ 23.98 fps Updated Corrected Horizontal Centering

I can't get this work on my 5d markii, has someone got it to work?

2blackbar

Dave can You upload Your 5D2 build with source? Id like to tinker a bit with CMOS values for centering .

reddeercity

Quote from: Macke1 on November 22, 2019, 08:07:21 PM
New Update: November 14/2019
3008x1080 @ 23.98 fps Updated Corrected Horizontal Centering
I can't get this work on my 5d markii, has someone got it to work?
All builds that I post are tested on my 5D2 & 50D , I don't post all thing that doesn't work or hasn't been test first by me
and I'm the one that coded it .

You need to explain you're self better , maybe some screen shots etc. .....
I can't help you if you don't help me by explaining the problem in detail .

reddeercity

Quote from: SKecorius on November 18, 2019, 04:26:11 PM
Thanks a lot for all your time and efforts. ......
I would like to ask, .....
#1) yes 10bit  and maybe 12bit
#2) Yes in 3xcrop_mode , in 3x3 you can record 10bit 1856x1248 @23.976fps -- there no 1920x1080 in 3x3 or FHD (Full HD)
#3) there is no moire and aliasing in crop_mode it's 1:1 pixel no binding or line skipping . 3k is just a extension of 3x crop_mode.
#4) no , dual iso has been working on video for the last few years , I have a build on the first post of this thread that works with
all the crop_rec & non crop_rec video modes but it will not work on the nightly builds just my builds work with it  dual_iso_video_5D2.mo

#a) I use 64GB but I hear 128GB cards work on the 5D2
#b) I don't think it will ever make the nightly , it still needs lossless compression implemented (it's close but missing something to be full functional)
More then likely the preset crop_rec will be obsolete soon , it will be replace with a resolution base interface , meaning you can
select any resolution and it will pick the max frame rate for that size e.g. 5632x3744 =5fps or 3008x1080 = 24fps etc. .....   
#c) No , I have not plan to maintain the basic feature set unless there something I need for me (I shot news stories , short docs & the odd low budget feature films)
I only have plans to implement Lossless compression & CF Card interface over clocking ( for higher write speeds)
#d) yes , read the first post of this thread

reddeercity

On 50D got vertical res working in 3xcrop_mode more then the base 1078 , I can go up to 1950
so I did a test and I can now get 3136x1302 @ 24.003fps , but the bad new is if it's like the 5D2 (adtg reg's hooks don't work)
so I guess the next step is trying to get the crop_rec module to build for the 50D so I can code these preset in to it


3136x1302_24fps_M23-0028_000000.png



So the adtg1[1172] & adtg1[1178] are the key regs to get larger vertical resolution then liveview e.g. 1067 or 1078

zcream

Quote from: reddeercity on November 23, 2019, 07:30:09 AM
#1) yes 10bit  and maybe 12bit
#2) Yes in 3xcrop_mode , in 3x3 you can record 10bit 1856x1248 @23.976fps -- there no 1920x1080 in 3x3 or FHD (Full HD)
#3) there is no moire and aliasing in crop_mode it's 1:1 pixel no binding or line skipping . 3k is just a extension of 3x crop_mode.
#4) no , dual iso has been working on video for the last few years , I have a build on the first post of this thread that works with
all the crop_rec & non crop_rec video modes but it will not work on the nightly builds just my builds work with it  dual_iso_video_5D2.mo

#a) I use 64GB but I hear 128GB cards work on the 5D2
#b) I don't think it will ever make the nightly , it still needs lossless compression implemented (it's close but missing something to be full functional)
More then likely the preset crop_rec will be obsolete soon , it will be replace with a resolution base interface , meaning you can
select any resolution and it will pick the max frame rate for that size e.g. 5632x3744 =5fps or 3008x1080 = 24fps etc. .....   
#c) No , I have not plan to maintain the basic feature set unless there something I need for me (I shot news stories , short docs & the odd low budget feature films)
I only have plans to implement Lossless compression & CF Card interface over clocking ( for higher write speeds)
#d) yes , read the first post of this thread
#4b Is the current 10-bit and 12-bit video in 5d2 achieved without lossless compression? By truncated lower bits?

Sent from my Lenovo TB-8703F using Tapatalk


alexboum

Quote from: zcream on November 23, 2019, 11:31:34 AM
#4b Is the current 10-bit and 12-bit video in 5d2 achieved without lossless compression? By truncated lower bits?

Sent from my Lenovo TB-8703F using Tapatalk

Yes read this post dedicated to the 10/12bit développement:
https://www.magiclantern.fm/forum/index.php?topic=5601.0

reddeercity

Looking like I'm very close to 1x3 (1376x1776) on the 50D some of the edges are a little jagged yet ,
my need a some more adtg_reg's like "adtg1[100c]"
but here's the first sample for the pixel peeper  :D


4128x1776_24fps_M25-0243_frame_3.png


1376x1776_M25-0243_frame_0.png

If any one what to try to get the same results here the reg's (note it takes time , and can't be rushed , adjust reg very slooooowly )  ;)


Edit: this is the max resolution for 1x3 @ 24fps ,
thou it is continuous recording @75MB/s , good that the 50D can record at 80MB/s  8)

reddeercity

Found the Liveview preview reg's for the 50D in 3x3 for 1x3 (1376x1777 @ 24.010fps)
(for the 50D you don't have to be in 3xcrop for it to work)


C0f08184 = Height
c0f08188 = Width
C0f0d014 = Height & Width

the only difference from the 5D2 is "c0f08518" is = to raw height & width where as "c0f0d014" is the reg in 50D

I came upon it by chance  :) , just look for the height & width in hex ( that is the total raw area including the offset)
c0f0d014 = 0x43b0680 (default 3x3)
0x43b = 1083 (Decimal Height Value)
0x680 = 1664 (Decimal Width Value)

So to do the math ,
Lv height in 3x3 is 1056+27 offset = 1083
Lv width in 3x3 is 1591+73 offset =  1664
and mlv_rec uses mod8 rules so the closes we can get is 1568 .

I connected my Atomos Ninja hdmi recorder to the 50D with 1x3(1376x1777)
Got a good clean(mostly  :D) footage , wanted to see if the HD VRAM is good to monitor or capture .
Some work still needs to done , the early results look really good though some of the raw footage is getting corrupted  .
so I need to find out why to use this .

In Standby Mode

footage from Ninja_hdmi_50D_720P_00240.mp4 ,

While record raw @ 1376x1777 24.010 fps

originally it was 1080i (8bit 4.2.2 uncompressed) so I did a reverse telecine form 3-2 pull-down (extracted the 24p frames from 60i stream)
then I resized it to 720p for a small file size conveniences .


zcream

Quote from: reddeercity on November 25, 2019, 08:49:42 AM
Looking like I'm very close to 1x3 (1376x1776) on the 50D some of the edges are a little jagged yet ,
my need a some more adtg_reg's like "adtg1[100c]"
but here's the first sample for the pixel peeper  :D


4128x1776_24fps_M25-0243_frame_3.png


1376x1776_M25-0243_frame_0.png

If any one what to try to get the same results here the reg's (note it takes time , and can't be rushed , adjust reg very slooooowly )  ;)


Edit: this is the max resolution for 1x3 @ 24fps ,
thou it is continuous recording @75MB/s , good that the 50D can record at 80MB/s  8)
This is great news. The Resolution is closer to 2.7k and the human eye is less sensitive to reducing horizontal resolution.
We used to have 1440x1080 and not many could discern the difference with 1920x1080

Sent from my Lenovo TB-8703F using Tapatalk


reddeercity

Quote from: zcream on November 27, 2019, 09:35:52 PM
This is great news. The Resolution is closer to 2.7k ......
Close , it really = to FF(full frame) 3.2K (it that's what you are referring to)
The crop factor on the 50D is 1.6 of FF as everyone know (APS-C) ,
with the 1x3 -> 4124x1776 the crop factor = to 1.15 of the full 50D sensor (4752 width)
so we get a crop factor of 1.75 of FF (full frame)  , 5D2 sensor = 5632 width (FF)
so 5632/1.75 = 3218

zcream

In practice 1376 px out of 4128 px columns will give about 2.7k resolution (as row resolution is unchanged). However as you are squashing horizontal it is less noticeable.
It's great news however you look at it. This is closest to 4k dci we can have.

Sent from my Redmi 4A using Tapatalk