Canon 100D / SL1

Started by nikfreak, October 19, 2015, 10:41:29 PM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

OlRivrRat

                     @IDA-ML

           Does this change @All with Lens or SD Card swaps?
ORR~DeanB  ~~  80D-ML  &  SL1+ML  &  5D2+ML  &  5DC+ML  &  70D+ML(AliveAgain)

dfort

Hum--I was in the unified branch when I looked up the skip values and it seems that Matthew Kelly's commit for the 700D has already been applied to the other cameras in the crop_rec_4k branch. Edited my previous post.

[EDIT] Shoot -- I made the pull request that made that happen. No wonder it looked familiar. Must have had a senior moment when I wrote that post.

I couldn't reproduce the "hiccup" issue on the EOSM or 700D.

IDA_ML

Quote from: OlRivrRat on January 20, 2018, 08:24:17 PM
                     @IDA-ML

           Does this change @All with Lens or SD Card swaps?

I tested with at least three different SD-cards, including a 95MB/s 64 GB Sandisk one, and with 5 different lenses.  The hiccup issue is always there.

As far as I remember, quite a while ago, Danne confirmed this behavior on his 100D too but I cannot find that post.  Maybe, someone else with a 100D could test too.  Canneloni, could you give it a try?

dfort

Quote from: IDA_ML on January 20, 2018, 06:13:54 PM
...high-quality video recording is possible at resolutions up to 2520x1080....

Just realized--how are you getting that resolution? The zoom mode file @canneloni did is 1920x1080. The maximum horizontal resolution I'm seeing is 2520 and the max vertical is 1078 using mlv_lite:

EOSM = 2520x1072, 1920x1078
700D = 2520x1072, 1920x1078

I'm not sure how the update_resolution_params function in mlv_lite works but it seems to me that the 100D should have the same combination of image sizes as the EOSM and 700D.

OlRivrRat

      @IDA_ML

   I would like to test but am unable to Rec' any more than about 2Sec's of Vid no matter what I try.

I'm guessing I'm doing something incorrectly. Could You Please post info on What Build You are using &

a Detailed Description of All SetUp Info. An Upload of Your ML/Settings Folder to somewhere could be

very helpful. Thanks ~

                                 ORR ~ DeanB
ORR~DeanB  ~~  80D-ML  &  SL1+ML  &  5D2+ML  &  5DC+ML  &  70D+ML(AliveAgain)

dfort

Quote from: dfort on January 20, 2018, 09:56:14 PM
how are you getting that resolution?

Getting closer to answering this question.

On the 100D zoom mode files the full raw buffer is a different height by 1-pixel and because the hight needs to be rounded to an even number the image size is different by 2-pixels.

100D zoom mode
Block: RAWI
...
    Res:  1920x1080 <-- image size
...
      height           1108 <-- full raw buffer
      width            2592


EOSM zoom mode
Block: RAWI
...
    Res:  1920x1078 <-- image size
...
      height           1107 <-- full raw buffer
      width            2592


I'm not sure if this is the cause of the "hiccup" issue but if I have a choice I'd rather have 1920x1080 than 1920x1078.

By the way, what I thought might have been happening was something that was discussed in this comment from the pull request that unified the 100D/650D/700D/EOSM skip offsets but all these resolutions are valid:

In [1]: 1920*14//8*1078 % 16
Out[1]: 0

In [2]: 1920*14//8*1080 % 16
Out[2]: 0

In [3]: 2520*14//8*1072 % 16
Out[3]: 0

In [4]: 2520*14//8*1080 % 16
Out[4]: 0


[EDIT] If you're wondering what's going on with resolutions between 2520x1072 and 2520x1080:

In [5]: 2520*14//8*1074 % 16
Out[5]: 4

In [6]: 2520*14//8*1076 % 16
Out[6]: 8

In [7]: 2520*14//8*1078 % 16
Out[7]: 12


These are not valid.

Danne

I´d say the issue is about getting identical frames repeated at random places. When the repeated frames is produced it will skip a frame so it's not like we are getting extra stuff here:
https://www.magiclantern.fm/forum/index.php?topic=16040.msg191480#msg191480

If you follow my links there are some additional info about some related issues where the next recorded MLV file would contain a frame from the preceeding MLV file. Guess there´s some intricate stuff going on with memory allocation(now that´s only rambling on my part  :P)

IDA_ML

Quote from: OlRivrRat on January 22, 2018, 12:59:21 AM
      @IDA_ML

   I would like to test but am unable to Rec' any more than about 2Sec's of Vid no matter what I try.

I'm guessing I'm doing something incorrectly. Could You Please post info on What Build You are using &

a Detailed Description of All SetUp Info. An Upload of Your ML/Settings Folder to somewhere could be

very helpful. Thanks ~

                                 ORR ~ DeanB

OlRivrRat,

The hiccup issue occurs in every 100D build with RAW video, even in the eldest ones.  I suggest that you use the Dec. 19-th build that you can find on the Experimental page.  In my post #935 on page 38:

https://www.magiclantern.fm/forum/index.php?topic=16040.925

I have provided detailed explanation on how the 100D behaves with that build in the different camera settings, including the problematic 5x-magnification mode.  To get more than 2s. recording times in that mode, please select the RAW-video mode, set FPS override to 24 fps, press the magnification button until you enter the 5x-magnification mode, set Preview to AUTO, chose a desired resolution and set the bitrate to 11...8 bit lossless which will give you the longest recording time.   Also please make sure that:

1) Your SD-card provides at least 45 MB/s write speed (run a benchmark test to verify this);

2) You have not selected 14-bit (uncompressed) at the highest 2520x1080 resolution - at this setting, there is no way that you get more than 2s. recorded.  That is the reason why I suggested  11...8 bit lossless;

3) Make sure, you do not overexpose your shot.  I typically first set the correct exposure (E0.0 reading on the histogram) and then I dial in one stop underexposure (E1.0).  This typically doubles my recording times with a well lit scene.

With the above settings, even at the highest 2520x1080 resolution, you should be able to get at least 6s. recording time.  If you don't, this means that your card is too slow.  In that case, please reduce the resolution to say 1728x972.  Hiccups occur at any resolution.

Finally, I would suggest that you shoot several clips at the above settings of scenes that contain moving objects (passing cars, running dogs, etc.).  Then, please convert the clips into H.264 files, (I typically use MLV Producer for that), and watch them carefully during playback.  You should see the hiccup jumps on the moving objects at random locations of the scene while the static ones stand still. 

And please report what you find.


IDA_ML

Dfort and Danne,

I keep asking myself why the hiccup phenomenon occurs only on the 100D and not on other cameras like the 700D or EOS-M which are almost identical to the 100D.  If the differences are analyzed, then the reason why hiccups are happening on the 100D may be more easily found.

canneloni

I will try to do a full test soon but just came across this idea: Did anyone try it with mv720 and fps override? I just tried it very quick and the file came out almost fine. Maybe there is a chance in seeing what happens there and if there is any significant difference ?
100D.100B ; Canon 18-55 STM ; Canon 50 1,8 II ; Canon 75-300 4,0 - 5,6 III ; Sigma 17-50 2,8

DeafEyeJedi

I can also confirm that the phenomenon hiccup horror does occur from recording in 5x zoom mode while in mv1080 (2520x1080p) for about 5-6 seconds with a fast SD card (FPS override enabled to 24p) using the latest Dec 19th build from experimental page. Previewing footage upon via MLRV 1.5.0 which clearly shows it as well as on exported footage.

Perhaps I am gonna go ahead and investigate on whether or not it is having similar identical pattern effect from one another when the 'hiccup' occurs based on the DNG's number in order. Just a theory I'm throwing out there on the table.
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

dfort

Uploaded a test build to my Bitbucket downloads page. Look for the "Hiccup Test" file.

What I did was the same as was done for the 5D3:

raw.c
#ifdef CONFIG_100D
    (*height)--;
#endif


How did I come up with this?

raw.c
    /* height may be a little different; 5D3 needs to subtract 1,
     * EOS M needs to add 1, 100D usually gives exact value
     * is it really important to have exact height?
     * for some raw types, yes! */


It looks like 1 wasn't added to the EOSM so I tried it and was able to get the full 2520x1080 resolution out of zoom mode but it had a "hiccup" at about the same rate as the 100D examples. It wasn't exactly the same kind of "hiccup" but let's try it out. If it fixes the "hiccup" issue maybe it could be refined so that it is only applies to zoom mode. Then again if it doesn't affect the other video modes we could consolidate the 100D with the 5D3.

canneloni

Just did a quick test with the newest build and recorded some video playing on my monitor in 5x Zoom and didn't see any hiccups. But i got different resolution options, now i got 1920x1078 . I also have (probably even before the latest build) problems with my SD Card which i have to figure out first to do real test with real recording times and resolution, but so far I didn't see any hiccup :) I used fps override to go from 30fps to 24.
100D.100B ; Canon 18-55 STM ; Canon 50 1,8 II ; Canon 75-300 4,0 - 5,6 III ; Sigma 17-50 2,8

OlRivrRat

      @IDA_ML & WMBI

   Was able to get a Short Quick&Dirty test run last night just doing a Pan @ My Computer.

2 Hiccups occurred & with Processing My Suspicions were confirmed > The Hiccup is a BackStep

& No Frames were Lost. Looking @ the resulting DNGs shows a Frame Sequence like this >

A - B - C - E - D - E - F - G

If that holds true on additional testing then if the Issue can not be resolved @Least the Vid's

Could be Rescued in Post.

                     This test was done with the Dec19 Build ~
ORR~DeanB  ~~  80D-ML  &  SL1+ML  &  5D2+ML  &  5DC+ML  &  70D+ML(AliveAgain)

Danne

Quote from: OlRivrRat on January 23, 2018, 04:39:59 PM
No Frames were Lost. Looking @ the resulting DNGs shows a Frame Sequence like this >
A - B - C - E - D - E - F - G

Could you do more tests? My experience shows that frames are missing.

IDA_ML

Quote from: dfort on January 23, 2018, 05:16:33 AM
Uploaded a test build to my Bitbucket downloads page. Look for the "Hiccup Test" file.

It looks like 1 wasn't added to the EOSM so I tried it and was able to get the full 2520x1080 resolution out of zoom mode but it had a "hiccup" at about the same rate as the 100D examples. It wasn't exactly the same kind of "hiccup" but let's try it out. If it fixes the "hiccup" issue maybe it could be refined so that it is only applies to zoom mode. Then again if it doesn't affect the other video modes we could consolidate the 100D with the 5D3.

This is very good news, Dfort!  I will test as soon as I get back from work.

Maybe, other owners of the 100D could test that new Hiccup build at different settings (FPS, bit rate, resolution) and share their findings too.

dfort

Quote from: Danne on January 23, 2018, 04:43:55 PM
Could you do more tests? My experience shows that frames are missing.

We know there's a problem with the Dec19 build. Try the Hiccup_Test.2018Jan22.100D101 build I posted on my Bitbucket downloads page.

DeafEyeJedi

Quote from: dfort on January 23, 2018, 05:16:33 AM
It wasn't exactly the same kind of "hiccup" but let's try it out. If it fixes the "hiccup" issue maybe it could be refined so that it is only applies to zoom mode.

Glad you brought this up. Because I was wondering the same few days ago when playing around with the Dec 19th build and thought it wasn't exactly the same type of hiccup so to speak but still very similar. Nonetheless, I went ahead and tried the latest Jan 23rd build (Thanks @dfort) and notice a few things.

•Seems you can't select 1080p for the height (seems to jump in between 1072, 1076 or 1078) while in 5x zoom with the Jan 23rd build but can with Dec 19th build.

Here's what it looks like on 100D LCD (recorded w iPhone since I don't have the Shogun presented w me) of what I was referring to from earlier...



•Upon rendering the MLV files via Switch (Thanks @Danne for the recent update) and haven't seen any 'hiccups' in both 5x zoom mv1080 & normal mv1080 as well as mv720 with exported ProRes files as well previewing them in MLRV 1.5.0 in prior.

Now I am not sure where to think of the reason why we're sacrificing few pixel lines just to avoid certain hiccups?

May have to investigate further more.

Oh yeah and Happy Birthday to the great @dfort!
5D3.113 | 5D3.123 | EOSM.203 | 7D.203 | 70D.112 | 100D.101 | EOSM2.* | 50D.109

dfort

Quote from: DeafEyeJedi on January 24, 2018, 02:55:40 AM
•Seems you can't select 1080p for the height (seems to jump in between 1072, 1076 or 1078) while in 5x zoom with the Jan 23rd build but can with Dec 19th build.

@DeafEyeJedi - Back up to Reply #1003 and start reading from there. It looks like you have a choice between 2520x1080 with hiccups or 2520x1072 without hiccups. Check the other video modes on the Dec12 build and my Jan23 hiccup build and see if other video modes are also affected. I would assume they are. @IDA_ML also did a test and found that Movie Crop Mode isn't working with the uncompressed 10/12/14bit settings. Try my hiccup build with those settings too and see if it works.

Danne

Seems to me you fixed the issue with doubled frames(hiccups) in 5xzoom mode dfort. That is very nice now that we have lossless settings allowing for longer recording times. Tested lossless modes 8-11, 10, 12, 14 bit and regular 10,12,14 bits

Where can we find your changes?
Nevermind. Checked a few posts up.


IDA_ML

Amazing work, Dfort!  I don't know how to thank you for everything you are doing to turn the small and insignificant 100D into a high-quality RAW-video shooting beast!  I also thank Danne for adding the remarkable work of Erwin Houtsma - sound in the MLV-lite mode.

I can confirm that the hiccup issue has been fixed and the 5x-magnification mode provides very smooth and clean results now.  I am getting between 140 and 160 frames in that mode at 2496x1078 resolution at 24 fps at 8...11bit lossless, unfortunately without sound.  There was sound with the Oct. 23-rd, 2017 build that I was using before.

Here are a few things that remain to be fixed:

RAW-video mode:
===============

1) No sound at 5x-magnification;
2) No in-camera playback of clips recorded at 10 and 12-bits uncompressed - black screen, frame counter is working though;
3) 10, 12 and 14-bit uncompressed video recording in the Movie Crop mode is broken.  Camera records a few good frames in the beginning and then all frames are corrupt.

RAW-video (MLV) mode
====================

1) No sound gets recorded at any of the bit rates;
2) No in-camera playback of clips recorded at 10 and 12-bits uncompressed - black screen, frame counter is working though;
3) Camera hangs when switching between modes and bit rates.  A "BUSY" or "Threads failed to start" message occurs every time a setting is changed.  A battery pull and restarting camera several times is needed to get a clip recorded, no matter if you are in the normal uncropped, 5x-magnification or Movie Crop mode.

I have no time to continue the tests right now but if I find more bugs, I will report in this thread.

IDA_ML

I also have a request to Danne.  Could you please add to your latest Jan. 24-th build also the latest IS feature and some delayed trigger options that A1ex added to the latest lua_fix build (January 20-th) for the 100D?  See here:

https://www.magiclantern.fm/forum/index.php?topic=12523.1000

https://builds.magiclantern.fm/experiments.html

In this way we will have maximum functionality in one build.

Thank you in advance.





Danne

Not sure where to look for changes asked for but I think the better way would be to deal with implementing audio to mlv_lite in crop_rec_4k branch and move from there. Feel free to have go here, I´m a bit limited atm...

Danne

@IDA_ML
Here is a build for 100D with sound included along with the fix for the "hiccup" issue included into crop_rec_4k branch. Not tested so go nuts:
Build:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/magiclantern-Nightly.2018Jan24.100D101_crop_rec_4k_snd.zip

Commit:
https://bitbucket.org/Dannephoto/magic-lantern/commits/23b21dddd5467526a11638ff1ae6858e9390c946

Added mlv_lite.c and mlv_snd.c from ErwinH sound build into crop_rec_4k so not sure what will work or not. ErwinH should probably do a PR to start things going...