Messages

sounds a lot like a free lunch

in photography, full frame is always better than aps-c.
in movie, probably rolling shutter is greater in full frame. resolution wise, i doubt there is any eye catching difference between the modes.

better dr
better low light
better gradation
better color (subjective)

so, although 3.5k has workable dual iso, this does not compensate the full sensor's advantage.

Image quality wise, 5D3 ML 3.5k and 5.7k AR 2.4 24p 14 bit lossless 14 bit color depth modes are no less than R1MX 4.5k ws RedCode 42 (42 MB/s) compression ratio 7.5 24p.
For rendering time in Davinci Resolve 18, R3D is about 3 times faster than ML raw 4k, more so than ML raw 5.7k.
Now lens is the bottle neck, instead of camera. Don't blame your camera, blame your techniques.
I suspect that once 7D2 and 5D4 have ML raw ported, they may be at least equivalent to BMPCC 6K/pro, Red Dragon 6k, or even Alexa 35 (maybe dr not so).

ML Rocks!

Ah, that part.  Yes, they did say that, and if you read the following posts you will see people exploring it and get the answer to your question (10-bit lossless is lossless).

thanks, good to know. only arri and ml raw are uncompressed. i really don't like there is any cloudy thought here about the lossless compression algorithm. it is one of the best selling points of ml raw. ml rocks.

Lossless compression is lossless compression. Of course you are loosing information by using a lower bit depth.

some user tested that 10 bit uncompressed raw had better shadow than 10 bit lossless.

"In order to test, I recorded the same scene at 3 degrees of brightness (F2, F4, F11).
As you can see in the screenshots, varying brightness with Lossless changes the Data Rate, but the latter remains constant when using Uncompressed.

Recording 10bit ettr and lowering exposure in post will yield a cleaner image than 14bit recorded severely underexposed.
Also, 10bit Uncompressed has cleaner shadows than 10bit Lossless (No red tint)

You can see the data rates at various resolutions in the screenshots.
This could be particularly interesting for those using 1080p 3x3."

https://www.magiclantern.fm/forum/index.php?topic=26851.msg243544#msg243544

[MOD intervention: Full quote edited. Long post, please follow link]

For those who didn't know, it is possible to record Uncompressed (As opposed to Lossless) with some presets.

so 10 bit lossless is not truly lossless?

the more important and practical is if 7d2 can do uhd 60p lossless 14 bit or even 10 bit raw. this makes ml on par with other high end cinema cameras. many cameras claim they can do 4k or 6k 120p or 240p, but typically there is cropping or dual iso disabled. etc, the sensor readout is not fast enough. so 4k 60p raw is very practical.

( 3840 * ( 3840 / 2.5 ) * 23.976 * 14 ) / ( 8 * 1024 * 1024 ) = 236MB/s
  ^ width       a.r. ^   fps ^  bit dpth.^      ^---^--------^ bits to Bytes to KB to MB conversion

x0.4 in @WalterSchulz computation is the lossless compression ratio, which seems quite optimistic to me depending of the overall video exposure and the ISO (close to 100=less noise pattern=better compression): if you just increase it slightly to x0.5 then you get ~118MB/s, which sounds more realistic in real conditions with potential partial overexposures.

In my experience, wide angle shots with heavy foliage the ratio is about 0.7, which is the worst scenario.

5472 x (5472:2.5) x 24 x 14 x 0.4 / 8 =  192 MByte/s

216 in benchmark - 10 percent for overhead (est.) = 194.4 MByte

Close call.

If sd is overclocked to 240 mhz, maybe can add another 40 MB/s. 

Roughly 6k. 5.4k 1x1 14 bit color depth uncompressed raw is golden, even for the next several decades.

Fixed the task problem, which was embarrassingly simple in the end.  The OS defines a maximum number of tasks, and this cam happens to be quite close to that limit.  By coincidence, ML tasks were just below that limit before running the benchmarks, and so the second one triggered an assert.

The limit is defined during OS initialisation, and we control that during boot.  I've added an option to increase the task limit, which can be defined per cam:
https://github.com/reticulatedpines/magiclantern_simplified/commit/9df8de520ff38e47386404a10a262062cb7ab09e

Benchmarks look quite promising:
https://i.imgur.com/7bbLlNv.png

wow cf 135mb/s sd 80 mb/s.
if ml works on 7d2, it easily betters 5d3, can do 4k or even 6k 1x1.

To my eyes, 650d ml footage is even better than that from some famous log based video cameras. Bravo.
I suspect that none of the log based video cameras can compete with 650d/700d/100d/m ml crop mood. Except the Alexas with 12 bit prores4444 (internal actually 14 bit).
In other words, only raw video cameras may compete with ml crop mood cameras and 5d3.

Have you tried turning sound recording off before recording? The way Bilal stated it in the first post suggests that no sound will be recorded when using that mode, but for it to work, you actually need to turn sound recording off.

Thanks. I did not. I will try sound off.

You need to turn off sound recording (beside 240 MHz, Small hacks to All/More), I mentioned this in first post, also in my build walkthrough video.

Thank you, Bil. You are right. I had the sound on.




It seems L/R from customize buttons doesn't do anything on EOS M, beside you can already change shutter speed using the wheel (check the linked video at 9:00).



It's not available yet, there are old maps but might not fit for all of new crop mood presets. I answered here too with some info.

Just tested Crop mood's 1x1 full resolution 14 bit lossless fps = 3 mode on 650d. Basically, this mode is not working. The script windows pops up for no reason. Also, at 3 fps AR 2.4 the calculated data flow is supposed to be around 40-50 MB/s according to ML, yet after only 1 frame recording stops. This behavior is very similar to the old version before Crop Mood.
I am redeploying my 650d to a dedicated time lapse camera, so this mode is critical to me. I am sure many other MLers like the full res 1x1 mode too. Once trying some full res 1x1 footage, I guarantee more will be addicted to it.
Thanks a lot to Bil's great work!

Define "optimized".

10 bits is 10 bits, it may yield banding or not according to the scene, but I don't see why different modes should alter this fact. This is straight maths after all.

Of course would love to be proved wrong, always nice to to learn something new.

According to my experience, some of the modes on ML are not optimized. Theoretically they should behave better but not in practice. Sometimes the ML versions matter also.

So for me, I have to test my main modes in each new ML version. Not to take granted that previously this mode works, it should work in the latest ML version. Of course, the main modes need thorough stress tests before applying to field projects.

I don't know about the codes side, just from my experience using ML cameras.

1x1 3.3K I think. I suggest you have a look at the MLV files yourself, I included a download link.

Why unexpected? Shadows were heavily pulled here. If 10 bits were not banding, I don't know what could band.

Some of modes on ML are not optimized. Not all modes are equal. I used 3.5k 10 bit extensively before due to the low data stream at the time. Field footage is  very good.

[(Full quote removed by Mod)]

My humble contribution to the discussion:
..
(Full quote removed by Mod)

which video mode were you using on 5d3. 10 bit had significant color banding. really unexpected.

If speed is important, do not use Windows. Using Unix you get much higher export speed on same hardware. Our algorithms are the same for all supported OS.

Thanks a lot, Masc. Eventually I will build a linux video station to fasten the encoding.

Unscientific observation: after upgrading Windows 8.1 to Windows 10, MLV App encoding speed (prores4444) seems to be 2-5 times slower than before. No rigorous test was done, just feeling so.

ML's auto white balance is much better than Canon's stock auto white balance.
If the ML menu->white balance->auto adjust kevin and g/m can be assigned to a button, one button ML auto white balance is very helpful when changing scenes.

How that? No problem here. Is there a message? File size is a don't care for MLVApp.

Thanks for your reply, MASC.
I just found out that I used dnxhr and dnxhd, they are not working. Also maybe due to that the hard drive is almost full.
When I use ProRes422 in export setting, seems working, either with short or long clips.
Prbobaly ffmpeg has issues with avid's codecs? 
Anyway, I am going to stick to prores 422 or prores 422hq or prores4444. Good enough for me. Speed wise, seems they are good enough comparing to avid's codecs.

using crop mood on 650d, i can record single files with a size like 100 gbs. On 5d3 and old 5/25/2022ml 650d, big files are separated into 4gb series.
now importing 100 gb files into mlv app, they can be graded, but can not exported. small files have no problem.

Very good job, Bil. This way MLV App will get rid of all of the focus pixels.

I just found out that when changing small hacks from on to all, I can get the same level of data stream as the 5/25/22 build.

Now with sd overclocking to 240 mhz, sandisk extreme pro 170 mb/s 256 gb, on 650d

UHD 1X3 14 bit lossless AR 1.66 24p continuous orange about 80 mb/s, 10 ms idle.
5.2K 1X3 10 bit lossless AR 2.39 24p continuous orange about 80 mb/s, 10 ms idle.

Two thumbs up!

Thanks a lot, Bil.
I just tested full res 1x1, 3 fps, 90 mb/s, 50% idle, good live view, continuous green for 33 minutes. thumbs up.