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 - Digital Corpus

#1
Share Your Videos / Canon 7D MLV Tests
April 02, 2020, 03:25:33 PM
I'm not a film maker. I come from shooting stills. I've had some time recently to play around with RAW video a bit and this is some of what I've shot; I consider them technical tests. Given my background, I used RawTherapee for processing. Some are short clips, some aren't.













#2
Just wanted to extend my own thanks to this app. I come from shooting stills but have always been interested in some motion. ML has definitely had my attention since the first April first announcement introducing RAW video. Recently due to some abundance of time, I've been playing with it again and this app in conjunction with RawTherapee have been an interesting journey. Granted I've only used MLV App to export DNG files, but still, thank you for the work on it.
#3
Tragic Lantern / Re: 7D Raw Thread
October 17, 2013, 10:08:20 AM
a1ex,

I'm trying to track down the method of reproducing this, but I've had it happen several times, but even with FPS Override off, I have occasions that the shutter speed is locked when ML is loaded. I'm not sure if you've encountered this, but I'll delete my config file and move on from there.
#4
Tragic Lantern / Re: 7D Raw Thread
October 13, 2013, 09:15:00 AM
Quote from: man_bites_dog on October 10, 2013, 04:42:09 PM
Hi All

Long time lurker over the last couple of years. Thanks to all the devs for making this incredible hack - it truly is remarkable!

Don't know if this has been discussed or even possible, (and I have read a LOT of posts and didn't see anything), but are there any plans on creating a module that could write either ProRes or DNXHD to the CF card? I don't mind not having RAW controls and would love the option of having a 10-bit 4:2:2 Prores I could work with. Even maybe MJPEG (like the 1DC) would be good for me.

Cheers
Quote from: 1% on October 12, 2013, 11:06:36 PM
Heheh, prores? DNxHD?

Let me know when you have an implementation running on PII 233mhz real time, then we'll put it in ML.

What 1% is saying is a little more complex to those who don't understand codecs and the hardware inside portable devices..

From my understanding the DIGIC is a ASIC ARM processor that controls the majority of the image processing on the camera and is responsible for image output. A JPEG is a type of image file format where the data from the CMOS sensor is processed by the DIGIC and compressed in a number of ways, including lowering it to an 8-bit-per-channel bit depth.

The information going into the DIGIC is the raw image data. As is common on most cameras these days, this information doesn't need to be compressed down into a JPEG. In fact we have access as a still image and refer to it as a RAW file. Side note: nomenclature is such that we refer to a RAW image in caps, but I don't know why specifically.

Now since a RAW image is essentially unprocessed, it's very "easy" for us to recorder this information. the DIGIC has specific hard coded algorithms for decoding a RAW image and compressing it into a JPEG image, thus its hardware accelerated and doesn't require [much] software/code to produce.

However,  a format such as ProRes or DNXHD would have to be implemented purely in software since the compression algorithms are not in the hardware. Given it's relative performance of a Pentium II @ 233 MHz, our ARM chip is not capable of running such code/software capable of compressing the RAW feed into any requested format without taking a reeeaaalllyyy long time. For HD or near HD resolutions, it's extremely unlikely to get any of those formats codec to run at those resolutions and fast enough to chug away at 24-60 fps.

Now, our DIGICs have a H.264 encoder in hardware as well, but with limited options and we cannot really access many of the registers to control it. However, we can, thanks to ML, adjust bit rate and general compression quality. As such, you have h.264 for video or RAW and for still images we have a lossless 4:2:2 silent photo, RAW, and JPEG.
#5
Tragic Lantern / Re: 7D Raw Thread
October 08, 2013, 09:07:55 AM
Quote from: edingacic on October 08, 2013, 04:07:59 AM
So what is the conckusion on VAF and rec in raw does it work or it is too soft
A little soft, but it's very usable. I plan on doing some more tests in the near future to more accurately show that.
#6
Tragic Lantern / Re: 7D Raw Thread
October 04, 2013, 05:48:10 AM
Quote from: arrinkiiii on October 03, 2013, 11:45:16 PM
For what i know from my experience and also writing inglish the best i can...

-I recorded some raw footage in forest with tokina 11-16 2.8 (excellent for 5x)   wend i open the dng's on after effects i just didn't what to believe... wtf !!! I just got moire and aliasing on almost my all clips.

     1- My window in after effects with the moire/aliasing clip is 1/4 of my screen. Wend i went to full screen  the moire/aliasing just disappear at the least in almost of the clips. Shrink image you get this kind of aliasing/moire.

     2- Moire/aliasing will be showing in line skipping. In raw, like the name say, will be a lot more notice then in h.264
The reason why when you are re-sizing the frame window and you get more moire is because of the resampling algorithm. Given that your perceptive artifacts were greater was because that AE was most likely using linear/bi-linear sampling. If bicubic or lanczos sampling were used, your smaller image would exhibit fewer artifacts. A good example of this phenomenon is on wikipedia: http://en.wikipedia.org/wiki/Aliasing.
#7
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 11:09:30 PM
All of my links are fixed. Sorry, that I made them incorrectly. I'm used to other forums requiring the url to be encapsulated with double quotes.
#8
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 11:00:17 PM
Quote from: 1% on September 27, 2013, 10:52:42 PM
Benchmark in photo mode. I'm seeing no 100s, who has besides 5DIII?
When I went back to look a week or two ago, the screen shot wasn't up anymore  :(. Anyhow, I'll pick up one of those cards when funds permit.
#9
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 10:29:20 PM
I suppose it would be helpful to provide the .MLVs...
VAF-N.MLV: bzip'd tar (~522 MiB), zip (~552 MiB)
VAF-Y.MLV: bzip'd tar (~552 MiB), zip (~586 MiB)
#10
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 10:00:23 PM
I used britom's 12/09/2013 build of ML for recording this test.

I know I'm not active much on here and I wish I can contribute more so here is what I've done. I have the SanDisk UDMA 6/60 MB/sec,  16 GB card so I'm a little limited on card speeds but I do get a solid 50-55 MB/sec on it so it works well enough. I've been playing with the official RAW module and the bleeding edge one from britom and I have a VAF-7D  8) but no I am not using an NLE; just my Mac and my Linux server.  I recorded ~200 frames with and w/o the VAF-7D, critically focused both, to see how well the VAF works and how much fine detail resolution is lost.

These recordings were done about 2 minutes apart, roughly 200 frames each, on the Tokina 11-16 f/2.8 (first version of the lens) @ f/11, ISO 100, with a shutter of 1/48th w/o FPS override. I'm comfortable using this lens on my 7D and doing large prints so using it for 2 MP video isn't a concern. I intentionally overexposed by 2/3rds of a stop, processed the DNG's with RawTherapee with highly tweaked/customized settings to extract as much detail as I felt possible and up-res'ing them to 1920x1080. From testing with Photoshop and RawTherapee I determined it was best to 'bake' the sharpening(1) with RT and upsize it in the export instead of doing a separate resize & resharpen with Photoshop. I then passed the PNGs to mencoder (it doesn't play well with tiffs) and x264 to produce the following video tests.

Combined video: MOV, AVI
Without VAF: MOV, AVI
With VAF: MOV, AVI

VAF-N PNGs (704 MiB zip)
MD5: b5d7bf9af890e72cf277201e5f6bb51a
VAF-Y PNGs (670 MiB zip)
MD5: 75f81070184c41b56179d72f52a41cf7

Note 1: One possible pitfall of the test I did is that I apply the exact same settings to both shots. I previously pixel-peeped and fine tuned the detail and sharpening settings with frames using the VAF. As the non-VAF shot is inherently sharper, there is some minor over-sharpening.

Personal note: I own Photoshop CS5 and Lightroom 4. LR4 is a very nice piece of software and can produced stunning results with it's amazing tonal controls. However, it doesn't have the fine tune capability required to try and mask out or minimize the aliasing caused by pixel binning and line skipping. As such, I decided to used RawTherapee which has such controls. If you'd like to see what I've done with the fine details and you have RawTherapee, have a look at this file. In future, I think I'll export TIFFs and use LR4 for the effective "grading" until I decide to get a decent NLE.

To encode the PNGs with x264 via mencoder, I used the following settings:
mencoder mf://*.png -mf fps=23.976 -o ./VAF-Compare.avi -aspect 16:9 -msglevel all=0:vfilter=4:statusline=5 -ovc x264 -x264encopts crf=20:frameref=8:cabac:level_idc=51:subq=9:threads=0:nofast_pskip:keyint=72:keyint_min=12:rc_lookahead=48:deblock=-3,-3:partitions=all:me=umh:bframes=5:aq_strength=1.2:deadzone_inter=13:deadzone_intra=7:qcomp=0.8 -oac copy



Edit: The x264 settings are such that the video is compressed with a constant rate factor, which is analogous to a constant quality level.

To get .mov files I demuxed with mp4box (part of the gpack package)
mp4box -aviraw video <input file>.avi
and then muxed with mencoder since I've had problems with mp4box's muxing
mencoder <input>.h264 -o <output>.mov -of lavf -oac copy -ovc copy
#11
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 04:43:06 AM
I know most of those on here are playing with video and have any number of NLEs at their disposal, collectively that is. I'm just a very technical user who's been doing still photography for some time, though I've wanted to contribute to the development if I can. I happen to purchase a VAF-7D before a vacation last yer and I've not seen a RAW test with and without the VAF on our camera. I have the videos at home with a post ready to go, almost. I'll post it up in a few hours. Until then, here are two DNGs to PNGs that I then used to make the video. Details to come...

Without VAF
With VAF

Enjoy pixel peeping, though please keep in mind I minimized the aliased pixels during my RAW conversion process.
#12
Tragic Lantern / Re: 7D Raw Thread
September 27, 2013, 04:33:59 AM
Quote from: 1% on September 22, 2013, 07:51:07 PM
Max is 91, mlv_rec will get better for sure, but we're missing ~10MB somewhere.
Pardon my ignorance, but where did you get this number? I thought I recall seeing some ~100 MB/sec benchmarks on the 1000x cards. I understand the caveat of GD and other overhead associated with the features of ML.
#13
Tragic Lantern / Re: 7D Raw Thread
August 17, 2013, 03:47:31 AM
Some benchmarks on the 7D show write speeds of about 88 MB/sec. Reliably though, 80 MB/sec is what others are getting. Now, despite this, the silent pic mode isn't able to write out files at this speed on a regular basis. I don't know why specifically, wish I didn't have my embedded project taking up my time so I could help, but the silent pics are writing out slower than 80 MB/sec in some cases.

According to this, the data rates come out to (in MB/sec)
24 fps & 30 fps
1728 x 576 - 39.87,  49.83
1600 x 534 - 34.22,  42.78
1472 x 626 - 36.91,  46.14
1344 x 610 - 32.84,  41.05
1280 x 582 - 29.84,  37.30
1152 x 622 - 28.70,  35.88

60 fps
1728 x 308 - 53.29
1600 x 382 - 61.20
1472 x 420 - 61.91
1344 x 436 - 58.68
1280 x 486 - 62.29
1152 x 464 - 53.53


Though for higher resolutions, assumed 24 fps, we have (in MB/sec):
1736x1156 - 80.38
2520x1200 - 121.12


Now, if we assume 80 MB/sec our top end speed, @ 14 bits, gives us:
~1.997 MP pixels a frame @ 24 fps
~1.6 MP pixels a frame @ 30 fps
~0.8 MP pixels a frame @ 60 fps

For the following aspect ratios, assuming that we can achieve a continuous ~80 MB/sec +/- 0.5 MB/sec, the maximum frame sizes are:
FR      3:2         16:9         2.35:1
23.976   1728 x 1152   1892 x 1056   2176 x 920
29.97   1550 x 1032   1696 x 944   1944 x 824
59.94   1088 x 736   1192 x 672   1368 x 584

Just demonstrating theoretical in case anyone has the time to test and push the boundaries of our hardware.
#14
Feature Requests / Re: Better Image Quality
August 12, 2013, 07:59:48 AM
Quote from: groundlessfears on August 07, 2013, 05:09:15 PM
check this link http://snapsort.com/compare/Canon_EOS_550D-vs-Nikon_D90

read advantages of Nikon D90 and see line no 3. and then let me know what they mean by this picture quality.?
To reply or not to reply, that is the question...

Let me say this. My day job is a supervisor in a photography department where we have 45-60 Nikon D90 cameras in operation in a given day. I was lucky enough to be the predominant influence as to how these cameras are configured for getting the best, out-of-the-box image quality possible. That being said, we haven't needed to change that configuration after a few years of deployment.

Now guess what? I'd rather drop kick the camera into oblivion after chopping it up with a hatchet, feeding it to a wood chipper, and then milling it into a fine powder. I cannot stand the functionality, performance, or image quality from the camera.

As a full disclosure, I own a Canon 20D, 30D, and 7D with the Tokina 11-16, Canon 17-55 F/2.8 IS, Canon 70-200 F/4 non-IS. There are 2 features I wish I had from the Nikon firmware that have significant use and no equal on the aforementioned Canon cameras. However, to each their own and use the best tool for the job.

I've shot a few misc assignments on RAW with the D90 and I have less latitude than my 30D. The performance of the D90 is crippled when the battery percentage is at 60% or below which causes additional shutter lag when taking a photo. Despite being in Manual, you have to wait for the metering timeout in order to change the exposure after a half-shutter press, which is an asinine limitation as well. Furthermore, shutter speeds are supposed to be 'intuitive' where one goes from 2 seconds, 1.6", 1.3", 1". But when you change to fractional seconds the '1.3', '1.6', '2', and then '2.5' don't make sense to the average person whereas 0.8", 0. 6", 0.5", and 0.4" do make logical sense. Oh, and the body like to over expose by 1/3 of a stop under normal circumstance and 2/3rds if you have active d-lighting on at normal or higher. The latter of which we've compensated for, btw.


All of that said, I'll mirror the same sentiments that the others have mentioned previously:

  • buy glass
  • learn to take good pictures
  • the body doesn't matter that much.

I hate the D90, but it works for what we do and it's sufficient for the average Joe.
#15
Quote from: codypowers on August 05, 2013, 08:43:44 AM
I do live in SoCal, and was considering purchasing one.  It would be great to test drive one.  How effective is it in the raw files?
My presence here is just to follow where things are. I have an embedded solutions electronics project, day job, automotive work, and LifeĀ® that consume my time. As such I haven't had a chance to try RAW. Hit me up via a PM if you're serious about borrowing and we'll see what we can do.
#16
Quote from: crazyrunner33 on August 05, 2013, 05:24:28 AM
Try putting the sharpening to zero, ACR always pushes the sharpness amount to 25.  Are you seeing this pattern when pixel peeping?
Video in the 7D is done not just through line skipping but pixel binning as well according to the release information. This is what is responsible for the grid like pattern. You need the VAF-7D. If either of you were local to SoCal, I'd consider loaning it out to you as I own one.
#17
If I got off my arse and compiled a 7D copy from the repo, I'd host it. I have no time for extra projects though and this would fall into that category. Hosting isn't a problem when I've pushed through 1 TB in 3 weeks off my home connection and Verizon hasn't cared...
#18
Tragic Lantern / Re: Raw video on 50d and 40d
May 29, 2013, 09:49:35 AM
Ok, call me ignorant. I've followed the 5D3 thread, and I've searched this thread. Why is the resolution limited to 1592 x 1062? I noted another user stating that the resolution is bumped down to this when trying to select 1920x1080. Also, why is the default fps at 30?

In terms of pixels a second:
1592x1062*30 = 50,721,120
1920x1080*24 = 49,766,400

In other words, in terms of bandwidth, 1080p @ 24 fps is "easier". Again, call me ignorant :)

Aside from that, congrats to the achievement.
#19
Quote from: sharpmusic on May 25, 2013, 10:07:03 AM
Yes sir, video and audio data rates have always been measured in bits hence CBR (constant bit rate, and High bit rate)
It's not necessarily audio and video. Its more the simple detail that it's a data rate. It's reduced down to bit for terms of how the controllers/processors interact, i.e. on a bit level. Storage is where bytes will come into play. Yes, we are still dealing with video in this context with RAW, but our storage devices for this medium are listed in the "X" notation for speed and MB/sec as well. As such, it's simpler for cmparing a RAW speed to teh card speed by using MB/sec instead of Mbps.
#20
Quote from: sharpmusic on May 25, 2013, 09:53:12 AMExamples
24bit @ 1080i @ 30fps :24*1920*1080*30=1.49 Gbps.
24bit @ 1080p @ 60fps :24*1920*1080*60=2.98 Gbps.

1080i @ 60 fps == 1080p @ 30 fps, fyi. Interlaced means that there are half as many rows.
#21
Your error was in notation. Upper case 'B' is bytes, lower case 'b' is in bits.

g3gg0 does plan on returning back to the 7D's development, but his focus lies on the RAW video development and regular life (I assume). Right now ML is running off of the slave CPU in the 7D. As far as he can tell, now RAW stream touches the slave. So, for the time being, maybe until the end of time, we will not have RAW video. However, we will get a good copy of ML. In earlier tests I've recorded 220+ Mbps h.264 video with CRB 20 on my Samsung 16 GB cards.
#22
Quote from: feureau on May 23, 2013, 04:08:53 PM
Although the 7D can record to UDMA7 cards, it seems the speed is limited to UDMA6 cards.

The 1000x cards on 5D3 can write at about 88MB/s, while on the 7d it tops out at about 66MB/s. A tad below the minimum needed for 1080p raw, it seems....

I'm seriously thinking of selling my 7D and get the 5D3 for the raw video...
True, however you should consider this:
1920x1080 @ 24 fps --> 2,073,600 pixels, 14 bits @ 24 fps --> ~83.1 MB/sec
1920x900 @ 24 fps --> 1,728,000 pixels, 14 bits @ 24 fps --> ~69.2 MB/sec
1280x720 @ 60 fps --> 921,600 pixels, 14 bits @ 60 fps --> ~92.3 MB/sec

Drop 14 bit-depth to 12 bit-depth, which is about ~85.7%, and those numbers become:
~71.2, ~59.3, and ~79.1
This is dropping from 16,384 levels per channel to 4,096 levels per channel. In the real world, this is a negigable difference. Now, there will be a difference fropping from 12 bit to 10 bit, 4096 to 1024.

We can reduce the bandwidth a bit more by dropping the resolution and with how sharp and clean RAW is, we can upscale a touch to compensate. Not ideal, but if you have a clean image, there will not be any noticable difference. Let's try 1720x960:
1720x960 @ 24 fps --> 1,651,200 pixels, 12 bits @ 24 fps --> 56.7 MB/sec

This is a ~14% drop in bit-depth, ~11% drop in size, and voila! No one will notice the difference in resolution. You'll miss a focus pull more often than you will miss the resolution.

Now granted, the 7D's semi-unique architecture will make it insanely difficult to crack if we ever get RAW. But if it does, it will not be far behind the 5D3. It'll definately be better than the stock H.264.


Edit: We've had successful 1920x1280 recordings with the 5D3. This is a touch over 100 MB/sec
#23
Quote from: Digital Corpus on May 16, 2013, 05:48:49 AM...CF and SD cards are Flash memory. There is a bit of a principle of Flash memory that *will* affect  your write speeds, and though these memory cards are not what we consider SSD's, they have to abide by the same rules...
#24
Quote from: Claymore Pictures on May 16, 2013, 07:37:16 PM
I need some help. I'm new to the ML world and learning how this all works.

I have a 7D with the FW 2.0.3
I downloaded the 7D Alpha 2 and pulled the contents onto a CF card and then went through the steps to load it on the camera.
However while it's in the process of loading it seems to stop and gives me the screen that says:

"Ensure a lens is attached. (which there is) Press start start/stop button to activate movie shooting."

The Alpha 2 doesn't seem to have loaded on the camera, or if it did I'm misunderstanding how to access it.

What am I doing wrong? Am I in a wrong setting and am unaware?

Thanks.
Peter-John
I've had no problems loading Alpha 1 or 2 on my 7D. What lens do you have attached? If it is a non-cpu lens, in other words it does not report aperture, focal length, or focus to the camera, then the camera may not know you have a lens attached to it. I cannot speack from experience because I have no such lens.
#25
@platu, glad I can give some insight.

Quote from: vipmask on May 16, 2013, 11:41:27 PM
I'll try this new one. It frames fall!

It is on this I was able to record 1920 x 1080 25 fps and 1920 x 1280 25 fps :D
http://yadi.sk/d/0ogqyTkb4uUvL
1920x1280 @ 25 fps --> 2,457,600 pixels, 14 bpp @ 25 fps --> 102.5 MB/sec
If it was at 24 fps, you'd only need about 98.4 MB/sec.

As demonstrated by real world testing of 1080p RAW video, on would need a card that performs few extra percent, ~3%-5%, above this number in order to safely record the video without dropped frames.

Quote from: mageye on May 16, 2013, 10:29:11 PM
These are my little tests. The very short edits are because this is how long I can record at 1880x720. You may also notice some jumping too. Of course this is far from perfect.

http://www.youtube.com/watch?feature=player_detailpage&v=p9uZXUGDJSA

Recorded on the 5DMKII with DURACELL Pro Photo 600x UDMA 16GB CF card. I think the build I was using was the one from the 13th?

I have ordered a Komputerbay 1000x 32Gb and am hoping that it will improve the situation (reading the posts on here I am glad I chose the 32Gb one instead of the 128Gb)
1880x720 @ 24 fps --> 1,353,600 pixels, 14 bpp @ 24 fsp --> ~54.2 MB/sec
This speed coincides directly with your memory card. That KomputerBay one will do the trick.

Quote from: 1% on May 17, 2013, 03:06:58 AM
I feel like the refactoring lowered write speed somehow.. I'm getting less frames.... either that or the card is getting slower?

Now in photo mode non LV it takes all the memory and you have to turn it off before LV will start.
If you're still having write speed issues, try a low level or full format of the card to see if that helps.