Menu

Show posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Show posts Menu

Messages - lordgnagey

#1
Quote from: 1% on August 05, 2013, 04:59:06 AM
No slice its Eeko. Gop should work tho, flush rate works.

I'm assuming this means that GOP and Flush rate will (more than likely) work, but are not included in this build? What's more, is there a way to port Tragic Lantern to the 650d? Has that been attempted?
#2
1%

Pardon my ignorance but I don't see a GOP option in the bitrate menu. Is there some other way to alter the GOP?
#3
Nanomad, I'm going to re-post this because I'm sure it was lost in the 1.0.1 - 1.0.4 deluge.

Are there plans to implement GOP and Slice control? I've read that it is currently implemented in Tragic Lantern.
http://www.magiclantern.fm/forum/index.php?topic=6913.0

I realize you've been recompiling for the new firmware. I'm not expecting overnight miracles. However, has this been discussed as an option for the 650d? Are there any known issues of compatibility?

Considering that the 650d's write speed is nearly double that of the 600d it stands to reason that we could probably do GOP 1 and pretty high H264 quality without hitting the write limit on the cardbus.
#4
Are there plans to implement GOP and Slice control? I've read that they are testing it on the 600d.

Has that been discussed as an option for the 650d?
#5
Is there any work being done to alter GOP size or create ALL-I frames in h264?

It seems like raw is no longer a viable option because of the sensor, but high bitrate ALL-I could give us a serious boost in quality.
#6
Quote from: papkee on June 04, 2013, 05:39:52 AM
Okay, here's what I've got for RAW. A little test video made at my local pier.



Excuse the camerawork; forgot a tripod and had to warp stabilize the h*ll outta it.

Nice job, looks like you put a little time into this! What did you use to remove the af dots? I assume the color/luminance noise reduction in photoshop? What were the settings used?
#7
Quote from: satriani on May 31, 2013, 03:20:42 PM
Hey guys,
to be always up to date, use the "Daily compiled builds from the last ML source" or take a look to my signature. The compilation occurs automatically on my server, then tested on my camera and published.
I will try my best to upload every day. And please don't forget guys, always to report the bugs ;)

Hey man. I wanted to say I really appreciate the work! And I have tested out the May 31st build. Apologies for being so direct, but with the adjusted frame sizes 1472 width at 2.35 actually pushes us over the limit in terms of the buffer. It no longer allows for continuous recording, because of this I will be using the older build (with 1440) for my 24fps raw tests.

Is there a way to free up more space for alternative resolutions? Could we do this by removing the options above 1728? I think it would be beneficial considering that these higher res options are useless with this camera's capabilities. Personally I would like to see something like 960 1088 1152 1216 1280 1344 1408 1440 1456 1472 1536 1728 as width options to find what res that has enough cushion to allow for continuous recording. 1216 width would probably be the highest 16:9 with continuous recording.

I do understand that you were attempting to keep all resolutions within a multiple of 64 to maximize efficiency of RAW recording, but sadly it has cost us the highest 2.35 resolution that could record continuously at 24fps. And from my perspective, the speed improvements do not outweigh the option for doing longer takes. 1456 which is 22.75 x 64 might be a decent option, but I assume that 1440 will be a safer resolution to work with as neither is a direct multiple of 64.

That aside, I will say that I did test FPS override at 2.5 fps and I got a pretty nice 3.0x CBR h.264 sunrise TL. Now if only we could implement ALL-I frames... :P
#8

Quote from: xanta on May 31, 2013, 12:56:27 PM
my card speed is the same performance as your 95 on 650D, i tried your 1440*614 and there is no skipped frames too
but the recording may stop itself automatically and suddenly, not because of the buffer is full, the screen is off and the reflex is closed, i can here that click sound, could you understand what i mean? and when i dump the RAW file to the raw2dng.exe the cmd window flashed and showed me nothing, i mean the raw2dng couldn't process the suddenly-stopped raw file

You probably hit the card capacity or the file limit 4.29gb, I'd bet on the latter. Sad thing is I can't see any way of easily recovering that data. You have to watch your GB size with raw_rec. It's very early yet in its development.

But yes keep under that limit and you'll be fine.
#9

Quote from: xanta on May 31, 2013, 11:27:41 AM
what about the writing speed while recording? i am using a Sandisk Extreme 45MB/s card and the writing speed is 34-35MB/s, under 1280*720 frames were skipped at about 300th frame then it keeps skipping, but if i use a 1280*692(1.85:1) then no frames were skipped

Write speed was estimated at 35.2MB/s. It labored around 33-36 and usually didn't spike any higher. I really wish we could max out the 50MB/s write speed, or even come close to 45... That would add a world of new resolution options.

However 1440 : 614 at 2.35:1  Is an acceptable option in my eyes. It's viewable angle isn't quite 2-perf Techniscope however it is still a tad larger than M4/3.

M4/3 2.35 -  17.30mm x 7.36mm
1440raw 2.35 - 18.24mm x 7.75mm
Techniscope 2.33 - 22.00mm x 9.47mm
#10

Quote from: xanta on May 31, 2013, 10:25:55 AM
That one was great! i wonder what SDcard you used for this, and is there any frame skipped?
i have my own 650D tested but the res must be set to 1280*720 so that it will not skip frames too soon

another problem i have now is the strange red and blue dots on the picture as i posted before

I used a SanDisk Extreme Pro 95MB/s

Red and blue dots are part of the colorful apparition that I have been complaining about. They are burned into the image sadly, Nanomad is trying to clear these up. I have to use Color and Luminance noise reduction in photoshop which seriously degrades the detail of the image.

No frames skipped.
#11
https://vimeo.com/67355212

I had to compress it pretty heavily to get it to upload, but the ProRes 422 file is about 1 gig. I've put the ProRes file on Dropbox and it should be able to view later, I'll update this post with the link when finished.

I'm going to wait until we figure out whether or not the color apparition can be dealt with in camera before I do too much more with RAW video.
#12
Quote from: red5446 on May 31, 2013, 02:48:18 AM
looks like with a t4i it´s best to get a 45MB/s  card because with a 95MB/s you don´t get better performance at all.
anyways it will be cool to know if t5i has the same limitation.
i think at least 10MB/s more will get the job done for 1280x720

I'm sorry my first response was off, I misread your post. You're right, buying a UHS-I 95MB/s card is pointless if you are only using it on the T4i...

HOWEVER future UHS-I compliant devices will more than likely take advantage of the more than double speed increase from the 45 to the 95 cards. The prices aren't that different, so I say why not future proof your cards?
#13
Quote from: red5446 on May 30, 2013, 10:52:33 PM
if useful i have some crash logs:
currently using the build of satriani.
works ok. but be careful i got a semibrick or err (i dont remember what number i just pull of the battery and all is allright)

if i use 24fps over ride i can obtain 600+ frames of 1280x720. That´s 3x times what i get the normal way. But it crashes if i choose other any option than optimized for low light.

1crash2. crashes2 crashes
assert log for all crashes:
LOG000.LOG
Benchmarks with a sandisk 45MB/s


Please anyone with a 95MB/s make a benchmark a post it, will be helpful to see what buffer are better and if by any chance there is at least 10MB more of performance over a 45MBs :)


UHS-I only appears to be useful when UHS is usable by the cardbus...




I'm only getting 37.9MB/s MAX! I've tried this with and without Global Draw enabled and I get the same result.
:( It literally hits a cieling. what card do you use? Is it this one:

http://www.amazon.com/SanDisk-Extreme-Pro-SDHC-SDSDXP1-032G/dp/B004JJQN8G/ref=pd_sim_pc_1
#14

Quote from: Aborgh on May 30, 2013, 07:59:10 PM
How did you get rid of the red lines?

Color and Luminance noise reduction in photoshop. But they do appear to degrade the image when turned up too high. In the future uploads this apparition will probably be more pronounced, but only to retain detail in the resulting video.
#15
Quote from: Aborgh on May 30, 2013, 01:34:22 PM

How do you change the resolution on raw video?

By adjusting the raw_rec module. There is a section to select aspect ratio and another for choosing horizontal resolution.
#16
Here's an early test. It's a 1600 x 540 3:1 (if memory serves) and sorry it got downsampled to 720p (I hate QuickTime presets). Ill have a more interesting full 1080p stream later tonight. Plus plenty of reference DNGs.

As a reference this is California sunrise at around 7:30am... Very bright but you can see the details both inside and outside.

http://www.youtube.com/watch?v=bnkeaINBd4s&sns=em

Edit: I promise the next upload won't be as ugly. :P
#17
Well I'm at work right now. I can upload the videos and/or some reference raw when I get home.

#18
Hey guys, been testing Satriani's build and I have some notes...

2.35:1 1440 x 614 23.973fps recording with NO skipped frames at all...  That is, until I hit the 4.29 gb barrier. ;) This is with Overlays off and sound options disabled.

Other than the artifacts (which can be hidden/corrected to a degree) it's looking really great! Moire is a nightmare still, but the rainbow swirls can be fixed pretty easily. I've been processing the files into video just to get familiar with the workflow. Using Photoshop to grade and DaVinci Resolve for final polish and convert into 1080p ProRes.

This was done using 16gb UHS-I 95MB/s Sandisk Extreme cards.
#19
As a note... the 650d can do raw_rec. The max for 2.35:1 is 1440 x 614 continuous on my Sandisk Uhs-I 95MB/s (card bus maxes out around 40 MB/s).

However there is an issue where strange marks  are burned into the data, more than likely this is on account of the hybrid AF CMOS technology in the new sensor. Hopefully Nanomad (who is porting ML 2.3 to the T4i) will be able to get some clearer frames.

Everything can be found in the New Ports 650d thread.
#20
Just based on numbers 36.2 MB/s is still way higher than any h264 compression would allow in terms of bitrate. Sharpness would not decrease dramatically as individual pixels are still being written into the raw. We might lose some dynamic range and some color depth (again I'm just theorizing here).

However, with the sensor in the t4i I feel 8-bit high res is better than 14 / 12 bit low res any day of the week. Not to mention it should cut down on the number of unusuable Magenta codec glitches that continue to appear in single frames. (some people say this is caused by a type of image buffer overrun?)

To bring back the main point, Nanomad and the rest of the ML team have some serious work ahead of them just trying to code out the af point burn in and making 650d raw video into a stable module.
#21
Could we downsample the RAW to 8 - bit? Would that be too much for the CPU to handle in real time? It wouldn't be as glitzy as 10 / 12 bit... But most of us that have prior experience with canon cameras (I've been shooting DSLR since 2009 on the t2i) know that the sensor in this camera is only capable of so much. If I remember correctly the c100 shoots 8-bit hd video in avchd 4:2:0, so in theory it doesn't sound so bad. This is baring in mind the special 8.1 MP sensor on the c100.

In theory this would set raw at 8 - bit per pixel (I believe), it seems to me that would be much more likely to support (near or at) HD/2k sizes in a raw continous stream. My math figures it would be below 40 MB/s which would allow a cieling so that Global Draw can be enabled during recording. As a bonus this might offer some really high frame rates to 5D mark III RAW, and maybe open the feature up to older cameras.

I know we don't even have proof of concept for the 12 / 10 bit yet, and I know the 650d still burns strange AF artifacts into the images. So those will definitely be the first and most time consuming. But, can someone tell me if this is an option?

Maths:

1924 pixels by 820 pixels = 1577680 pixels per raw image
Which would work out to 8 - bits or 1 byte per pixel resulting in 1.5046 MB per frame or

36.2 MB/s for what could be 24 fps scaled at 1920 x 820 2.34 : 1
#22
I only use the SanDisk UHS-1 Cards, and I will absolutely be testing RAW .dng sequence video like crazy when it is implemented. I will probably do a test shoot in Death Valley or on Mt. Wilson just to put it through the paces.

I've successfully installed the second pre-alpha on my camera and I have noticed a bug, I'm not sure if it's been reported before:

When using the standard zoom function (magic zoom turned off) in Manual Video mode i noticed that I got a strange pink abberation of lines and static (in the box) whenever I moved the frame or tried to adjust the focus box. I'll retest when I get back to see if the issue persists.
#23
Long time magic lantern user here, (I was privately testing the beta on the 550d when that came out.) I wanted to say thank you to Nanomad and A1ex for devoting their time to such a worthwhile effort. Also thanks to blade, breaker, baga and all the testers for putting your cameras on the frontline to help with dev. I think that the final ML release for t4i is going to be a game-changer, just like it was for the 550d/MarkII back in the day.

Really excited for this Alpha!
#24
Very excited to test this build!