Shootout in extreme low light - 5D Mark III vs FS100 vs GH2

[Andrew Reid]

[html]

This is a torture test of the low light abilities of 3 cameras. There are three test scenes in a battered old ex-factory in the east of Berlin. The first consists of a completely pitch black room with glowing pearl lighting. The next one is in the roof is lit using a single iPhone 4S torch with the cameras ramped up to ISO 12,800. The third scene is again a pitch black dark room with the lights of a building outside reflecting in a mirror on the wall.

[url="http://www.eoshd.com/content/8062/shootout-in-low-light-5d-mark-iii-vs-fs100-vs-gh2/"]Read full article[/url]

[/html]

[jaescor]

Andrew,

Beautifully done. And a perfect song choice. What's is that song, btw?

Jae

[OlegKalyan]

Thanks for the test!

[palisady]

thanks for the test, but please NEVER use IRIS transition in test, specially when comparing how they do in the dark! By the time the compressed IRIS transition was over I totally forgot what the scene looked like and confused the hell out of me. Was the artifact there because the IRIS or was it the camera? I never recovered from this feeling for the rest of the test, which was probably interesting :)

[riogrande100]

Canon need to release a 5D Mark III without AA filter! And upgrade the firmware for codec and they have a much better camera!

Canon already released the 60Da for Astrophotography! Why not a 5Da considering a lot of people use the 5D for Astrophotgraphy and with the low light performance will reduce the shutter times!

I do expect that a 5D Mark III firmware update will probably update the codec! but the main blocked is the AA!

[richg101]

all looked stunning.  i was amazed how good the B+W bike shot looked with the GH2.  if anything the grain on the GH2 added a beneficial factor to the image.    all but the extreme 12,800 shots were really low in noise. 

what post processing are you applying to this footage?  any de-noise in software?  im never sure whether to leave the in camera de-noise on or off with the sony nex 5n.  is it worth spending money on neatvideo for CS4?  I cannot afford to upgrade to cs6 + aftereffects yet.  so the 50gbp neatvideo looks worth it..  any comments?

[gene_can_sing]

oppsss... I meant to post this here, but accidentally posted to the Vimeo page.

On thing I noticed on the Window test was that the FS100 had (what looked like to me but I could be wrong because I've only watched it once) a lot of internal reflections of the window (not good) off the sensor. The GH2 had it also to a lesser degree, and the 5D3, the least.

The good thing is, I saw an Abel Cine low light test between the FS100 and FS700, and the internal reflections were a lot less on the FS700 and the highlights looked considerably better.

But yeah, I was surprised by how well the 5D3 kept up with the FS100, but the FS still seemed a bit cleaner.

I watched about 1000 vimeo videos this weekend since I'm debating between the 5D3 or the Sony FS camera. At first, I was really biased towards Canon, but the more videos I saw, the more I started leaning towards the Sony. The FS cameras has the better resolution and the aesthetics are somewhere between the Canons and the GH2.

It seems like the FS cameras have a bit more dynamic range than both the Canon and the GH2 (I could be wrong). I was a big Shallow DOF fan, until I saw the Sony image. It actually felt a bit better because the blur was less overall. The 5D3 felt almost like it had too much blur a lot of times, something I never thought I would say since I love shallow DOF.
My friend who is a huge 5D fan, also felt the same way about the FS cameras. He also started leaning towards the Sonys by the end of the weekend, saying that the image just felt much more solid, more RED like, and it still had a nice aesthetic to it.

Strange how you mind changes

[Andrew Reid]

[quote author=richg101 link=topic=724.msg5327#msg5327 date=1337017730]what post processing are you applying to this footage?  any de-noise in software?[/quote]

None. Straight off the card and onto Vimeo. No fluffing.

[see ya]

[quote author=Andrew Reid link=topic=724.msg5332#msg5332 date=1337025294]
[quote author=richg101 link=topic=724.msg5327#msg5327 date=1337017730]what post processing are you applying to this footage?  any de-noise in software?[/quote]

None. Straight off the card and onto Vimeo.
[/quote]

Except the 5D MkIII full range levels have been squeezed into 16 - 235 skewing results, where as the FS100 full luma levels have been left as shot in camera and for whatever reason the GH2 levels are 16 - 235 as well.

The 5D MKIII MOV as well as Nikons MOV's use h264 VUI Options metadata, including a fullrange flag set 'on' by the camera firmware, this forces the decompressing codec to squeeze luma levels, evident in the image extracts in the zip below, showing the luma waveform, notice the fine horizontal lines at regular intervals in the waveform on the 5D MKIII image, clearly showing luma levels have been compressed.

http://www.yellowspace.webspace.virginmedia.com/bike.zip

The images were extracted using AVISynth.

If the Canon MOV is remuxed with a h264 VUI Options patched build of MP4Box here: http://komisar.gin.by/tools/ setting the fullrange flag to 'off' in the process then the full levels as shot in camera are left well alone by the decompressing codec and the outcome would be very similar to the FS100 including less contrast, brighter image than the 5D MKIII is showing currently.

As an aside the GH2 levels in the attached image show restricted to 16 - 235 levels. I'm not aware the MTS container holding a fullrange flag and looking at the waveform in comparison to 5DKMIII there are no fine regular horizontal lines to suggest luma is getting squeezed, gradation looks 'fine' like the FS100 and what the 5D MKIII would look like if decompressed correctly, but GH2 looks to capture to restricted 8bit range?

A further minor skew is that Canon native files use BT601 luma coeffs where as the GH2 and FS100 are BT709 so there will be a slight skew in contrast / color of a Canon source due to the Vimeo mp4 being flagged as BT709 color matrix, which is incorrect for the Canon file unless a BT601 to BT709 color matrix transfer has been done?

[Andrew Reid]

5D3 and GH2 are 709 in-camera, not 601... As far as I am aware.

The FS100 picture profile settings have an option for 709 mode. It looks dreadful with it enabled.

I'll look into MP4Box, cheers.

[MattH]

[quote author=yellow link=topic=724.msg5335#msg5335 date=1337033035]
[quote author=Andrew Reid link=topic=724.msg5332#msg5332 date=1337025294]
[quote author=richg101 link=topic=724.msg5327#msg5327 date=1337017730]what post processing are you applying to this footage?  any de-noise in software?[/quote]

None. Straight off the card and onto Vimeo.
[/quote]

Except the 5D MkIII full range levels have been squeezed into 16 - 235 skewing results, where as the FS100 full luma levels have been left as shot in camera and for whatever reason the GH2 levels are 16 - 235 as well.

The 5D MKIII MOV as well as Nikons MOV's use h264 VUI Options metadata, including a fullrange flag set 'on' by the camera firmware, this forces the decompressing codec to squeeze luma levels, evident in the image extracts in the zip below, showing the luma waveform, notice the fine horizontal lines at regular intervals in the waveform on the 5D MKIII image, clearly showing luma levels have been compressed.

http://www.yellowspace.webspace.virginmedia.com/bike.zip

The images were extracted using AVISynth.

If the Canon MOV is remuxed with a h264 VUI Options patched build of MP4Box here: http://komisar.gin.by/tools/ setting the fullrange flag to 'off' in the process then the full levels as shot in camera are left well alone by the decompressing codec and the outcome would be very similar to the FS100 including less contrast, brighter image than the 5D MKIII is showing currently.

As an aside the GH2 levels in the attached image show restricted to 16 - 235 levels. I'm not aware the MTS container holding a fullrange flag and looking at the waveform in comparison to 5DKMIII there are no fine regular horizontal lines to suggest luma is getting squeezed, gradation looks 'fine' like the FS100 and what the 5D MKIII would look like if decompressed correctly, but GH2 looks to capture to restricted 8bit range?

A further minor skew is that Canon native files use BT601 luma coeffs where as the GH2 and FS100 are BT709 so there will be a slight skew in contrast / color of a Canon source due to the Vimeo mp4 being flagged as BT709 color matrix, which is incorrect for the Canon file unless a BT601 to BT709 color matrix transfer has been done?
[/quote]

Wow that is a hell of an information overload for one post.

After reading several times I still can't say I fully understand it.

So to get the full luma levels are you saying you have to do something before you record the video, or do you do something to the file afterwards?
Would you mind explaining how you achieve this?

Also if this can be done for the 5dmk3, can it be done for other canon dslrs?

[see ya]

[quote author=Andrew Reid link=topic=724.msg5336#msg5336 date=1337036645]
5D3 and GH2 are 709 in-camera, not 601... As far as I am aware.[/quote]

I don't know how long its going to take me to get this through my thick skull, yes Canon have changed to BT709 color matrix coeffs for the Mk III where as the MK II, 7D T2i etc use BT601. I even posted that the matrix had changed in comment at prolost in response to Dan Chungs test files back when the MK III came out, still forget. :-)

But the matrix is more to do with tweaking the color in specific hues pinks/orange blue/green and very little to do with luma levels, full range etc. All these cameras are BT709 color primaries defining the extent of the gamut in tge 1931 CIE color space horse shoe. But that is a separate thing altogether. Thanks for the correction. :-)

[quote]The FS100 picture profile settings have an option for 709 mode. It looks dreadful with it enabled.[/quote]

I think that probably relates to rec709 curve on the linear data rather than color matrix. Or perhaps its primaries. Not knowing the camera.

[quote]I'll look into MP4Box, cheers.[/quote]

Its a special patched build and also commandline. So to use it would be something like: mp4box -add my.MOV:fullrange=off -new my.mp4

The build doesn't like the sowt audio though but will give the necessary luma levels.

@MattH theres a bit here at my blog that may help

http://blendervse.wordpress.com/2012/04/02/waiving-the-fullrange-flag/

Although NLE specific the gist is the same.