15 replies to this topic

[echeng]

Our good friends over at Backscatter have just published an incredibly comprehensive review of the Canon 5D Mark II digital SLR's performance as an underwater video rig. The article opens with a sample video presented in 720p, followed by comparisons of the 5D Mark II to other professional / prosumer video cameras, a discussion of pros and cons, underwater housings, use when mounted on a scooter, color correction filters, tripods, and more. If you are even remotely interested in using any SLR to shoot video underwater, you should click over there and read it now!

LINK:
http://www.backscatter.com/5d2/

[wagsy]

Great work, wonderful review and shots.

[Jules (Helioxfilm)]

Nicely done, thanks! Anyone wants to buy my V1 Sony?

[Mike L]

Article is very comprehensive. Video is excellent as shot on the Dive X scooter.

[Elainew]

Thanks for posting this - will read with interest.

[bearaway]

Lovely dolphins video, always nice to see those guys

Regarding the review, it's brushing over the format and some comparisons with others, I find this a bit confusing and the author doesn't really detail his production workflow with H.264 either which is likely of prime importance for people that are interested to jump into it after reading the review.

The H.264 compression format is more efficient than HDV or MPEG and results in a better quality image at the same file size. This combined with a higher bit rate than the FX1 or EX1 results in less compression being applied, and therefore a higher quality image.

Those 2 sentences and the causality is a bit confusing here.

H.264 (inter-frame) is a nice end-user delivery format and gives on a general basis better quality for the same bit rate compared to MPEG-2...(and 40mbps is pretty high) however it is a very resource intensive format for both encoding and decoding... and is absolutely not appropriate for editing purposes. The playback of HD movie generally would put quite some stress on a dual-core cpu, so for playback you'll be better off having a codec that can take advantage of the GPU such as the nVIDIA CUDA. (we have used CoreAVC decoder on a previous job and if someone needs smooth playback on even single core, that is a well-optimized decoder and well worth $15)

As far as I know (and I'm not a movie person), if someone wants to do editing, he would first need to transcode H.264 using an intra-frame codec such as Apple ProRes, Cineform or DNxHD (one can do it in other formats...but I do think a pro would likely use those intermediate formats).

From my perspective, it would be interesting to know the workflow that was used in producing the clip from H.264 and what transcoding times and disk space usage are achieved before/after editing as it would give a better clue to people what are the implications in term of workflow (and hardware) of having to deal with the 5DM2 output.

[seagrant]

The review was very comprehensive and interesting, even for someone who mainly shoots stills!

Thanks! Carol

[Drew]

Bearaway
You are asking the right questions and the specs of the 5D2 are a little misleading. The bit rate is high but the sensor sampling is done rather poorly, resulting in aliasing artifacts which causes the Digic IV encoder to lose a lot of detail, even at such a high bitrate. With all the compression bandwidth the 40+mbps gives, junk in junk out, well not exactly junk but you get what I mean.
What does this all mean? The compression artifacts from the aliasing can be pretty bad with a sharp lens and deep DOF. I think one of the reasons why every loves the 5D2 is because the drawbacks of the sampling actually enhances the qualities which many people like, false detail that people see as "sharpness" when it's just aliasing. If you look carefully, there is less detail in the image. People with wrinkles love this camera.
That's the biggest reason why BBC won't even consider this a B cam even. However, there are other broadcasters who would, as can been seen by various TV and even one film project I know which is using the 5D2. The super sharp fast lenses, low noise and DOF control are the reasons why it's popular topside.
As for workflow, yes transcoding is needed to be efficient. ProRes, DNxHD, Cineform and Canopus all work the same way. A normal computer with 4-8 cores would do transcoding at 1-1.7x realtime using CPU with OpenGL GPU acceleration. nVIDIA CUDA GPUs like the FX4800 or Quadro CX go even faster, about 30-40% improvement in performance.
Hopefully in the October shootout of the 5D2 housings I'm planning, we'll cover some of the more technical stuff.

[TheRealDrew]

From my perspective, it would be interesting to know the workflow that was used in producing the clip from H.264 and what transcoding times and disk space usage are achieved before/after editing as it would give a better clue to people what are the implications in term of workflow (and hardware) of having to deal with the 5DM2 output.

Backscatter probably went with Final Cut for this clip since they did it before As described here Wagsy mentioned his workflow here on tests with 5D footage on Edius Thread

Not sure how far back things like this will work with computers that are older at a reasonable working rate, but with all these things like RED and the rest, the newer the better. (Non Intel Macs are pretty much obsolete for these things and the new Studio does not even run on them. I still use my DP G5 but go to the newer machines for heavy lifting). I transcode and use ProRes on my MacIntel Laptops and Octo Tower and it is fine. I will set up things to convert, run a batch job using all the cores and come back when it is done. Usually (based on time constraints and type of project) I will load things up to take of the transcode overnight. The Pro Res does take up some room, but with hard drives being pretty inexpensive the storage issue does not bother me that much, but the ProRes can take up good chunks of room.

I just pulled a a 22 second HDV clip and put it on my laptop and converted to ProRes (note it is usually better not to have media on laptop drive for better performance), took 32 seconds on a Dual Core 2.5, file went to 283 Megs from 198 Megs. Different material and different running times will change to some degree. I have not tried on 5D footage yet, but can run some tests at some point and let you know if you like.

The general descriptions of the ProRes flavors in terms of highest to lowest is (as described by Apple) :

The Apple ProRes 4444 codec is Full-resolution, mastering-quality 4:4:4:4 RGBA color (an online-quality codec for editing and finishing 4:4:4 material, such as that originating from Sony HDCAM SR or digital cinema cameras such as RED ONE, Thomson Viper FilmStream, and Panavision Genesis cameras). For 4:4:4 sources, a data rate that is roughly 50 percent higher than the data rate of Apple ProRes 422 (HQ)

The Apple ProRes 422 (HQ) codec provides target data rate of approximately 220 Mbps (1920 x 1080 at 60i)

Apple ProRes 422 (LT) roughly 70 percent of the data rate of Apple ProRes 422

The Apple ProRes 422 (Proxy) codec is intended for use in offline workflows and provides roughly 30 percent of the data rate of Apple ProRes 422

[bmyates]

Very nice review - makes me think I'd better start shooting video again seriously after all these years of shooting stills instead!!

[Mike L]

I cannot speak for other editing software, but just today I went through some of my clips of whale sharks shot on the 5d2 and edited them in FCP 6. All I had to do was import the clips as I do with my EX1, edit them as I wish, and then I just exported using Quicktime Conversion to H.264 and everything worked like a charm. I uploaded it right to Vimeo and then to Facebook without any problems.

[bearaway]

You are asking the right questions and the specs of the 5D2 are a little misleading. The bit rate is high but the sensor sampling is done rather poorly, resulting in aliasing artifacts which causes the Digic IV encoder to lose a lot of detail, even at such a high bitrate. With all the compression bandwidth the 40+mbps gives, junk in junk out, well not exactly junk but you get what I mean.
What does this all mean? The compression artifacts from the aliasing can be pretty bad with a sharp lens and deep DOF. I think one of the reasons why every loves the 5D2 is because the drawbacks of the sampling actually enhances the qualities which many people like, false detail that people see as "sharpness" when it's just aliasing. If you look carefully, there is less detail in the image. People with wrinkles love this camera.

As far as I know the DIGIC 4 does not use B-frames for H.264 encoding. This means it is effectively something like a Constrained Baseline Profile used for encoding... which is fairly logical since it is more widely supported... it of course also has the additional benefits of decrease processor complexity, keeping power consumption low and not having to require additional memory... the end-result of this is as you mention indeed that at the very least, it would likely miss out some texture details when there is dynamic/motion and it's a fairly complex background, but how much of this is theoretical vs practical and what are the cases were you miss out is probably hard to know (and would likely be a miss for 0.1% of the market targeted by this camera/video).

I would naively assume that in some way for underwater video with wide angle and fast motion like with the dolphins video, the large amount of blue water is very beneficial.
Even for macro there shouldn't be much issue, H.264 is amazing even at low bitrate such as 384-512kbps ('low-quality' web streaming) when something with shallow DOF (a nature documentary of fairly small animals)

Oddly enough I haven't seen any discussion related to those technical video details on all reviews I have read about it online (while pictures tend to be analyzed in/out in camera reviews), only a norvegian blog was mentioning the structure of the GOP for the 5DM2 but that is all that I have seen.

Hopefully in the October shootout of the 5D2 housings I'm planning, we'll cover some of the more technical stuff.

Excellent

[bearaway]

I cannot speak for other editing software, but just today I went through some of my clips of whale sharks shot on the 5d2 and edited them in FCP 6. All I had to do was import the clips as I do with my EX1, edit them as I wish, and then I just exported using Quicktime Conversion to H.264 and everything worked like a charm. I uploaded it right to Vimeo and then to Facebook without any problems.

AFAIK it would depend on what you do during the editing and whether it really needs to re-encode it. did you do some cut here and there with some transition ?
This is mentioned by MaryLinn in the thread that Drew pointed out related to the workflow for the 5DM2.

The next version of FCP is mentioned for 'realtime editing' of H.264 however
1) it makes it a somewhat not really critical feature because h.264 is really a delivery format
2) I suspect it is dependent of Snow Leopard as it will have OpenCL support and makes entire sense to go through it rather than do something specific. So until you have drivers and the relevant gfx cards, you're probably at the very minimum 6 months timeframe.

[JimDeck]

Hi Bearaway,

We converted the clips to Prores 422 in Final Cut. This made them easier to deal with. Editing in H264 is not really an option. I tried it and the bit rate drops for some reason when it is output. When I added titles, it dropped to 4 mbps.

I can playback the H264 at full res on my almost 2 year old MacBook Pro 2.4 core duo with 2 gigs of ram. When I output the 422 pro res file at full res, 179 mbps, it still played fine on my computer.

We'll have to see what editing in H264 will look like when it is included in Final Cut. This will help speed up the adoption of cameras like the 5D II in the future for journalists in mass media who don't have the time to convert to 422.

The review was getting pretty long with all that we tested. Editing in Final Cut is a whole other article/review in of itself.

Best regards,

Jim

Lovely dolphins video, always nice to see those guys

Regarding the review, it's brushing over the format and some comparisons with others, I find this a bit confusing and the author doesn't really detail his production workflow with H.264 either which is likely of prime importance for people that are interested to jump into it after reading the review.



Those 2 sentences and the causality is a bit confusing here.

H.264 (inter-frame) is a nice end-user delivery format and gives on a general basis better quality for the same bit rate compared to MPEG-2...(and 40mbps is pretty high) however it is a very resource intensive format for both encoding and decoding... and is absolutely not appropriate for editing purposes. The playback of HD movie generally would put quite some stress on a dual-core cpu, so for playback you'll be better off having a codec that can take advantage of the GPU such as the nVIDIA CUDA. (we have used CoreAVC decoder on a previous job and if someone needs smooth playback on even single core, that is a well-optimized decoder and well worth $15)

As far as I know (and I'm not a movie person), if someone wants to do editing, he would first need to transcode H.264 using an intra-frame codec such as Apple ProRes, Cineform or DNxHD (one can do it in other formats...but I do think a pro would likely use those intermediate formats).

From my perspective, it would be interesting to know the workflow that was used in producing the clip from H.264 and what transcoding times and disk space usage are achieved before/after editing as it would give a better clue to people what are the implications in term of workflow (and hardware) of having to deal with the 5DM2 output.

Edited by JimDeck, 10 August 2009 - 01:46 PM.

[Drew]

As far as I know the DIGIC 4 does not use B-frames for H.264 encoding. This means it is effectively something like a Constrained Baseline Profile used for encoding... which is fairly logical since it is more widely supported... it of course also has the additional benefits of decrease processor complexity, keeping power consumption low and not having to require additional memory... the end-result of this is as you mention indeed that at the very least, it would likely miss out some texture details when there is dynamic/motion and it's a fairly complex background, but how much of this is theoretical vs practical and what are the cases were you miss out is probably hard to know (and would likely be a miss for 0.1% of the market targeted by this camera/video).

I would naively assume that in some way for underwater video with wide angle and fast motion like with the dolphins video, the large amount of blue water is very beneficial.
Even for macro there shouldn't be much issue, H.264 is amazing even at low bitrate such as 384-512kbps ('low-quality' web streaming) when something with shallow DOF (a nature documentary of fairly small animals)

Oddly enough I haven't seen any discussion related to those technical video details on all reviews I have read about it online (while pictures tend to be analyzed in/out in camera reviews), only a norvegian blog was mentioning the structure of the GOP for the 5DM2 but that is all that I have seen.

Well the problem with getting too technical is that it goes beyond my own comprehension of sensors and sampling, and if I get bored, well I figure most everyone who don't live and sleep with a computer with a human name, would be too.
We first have to look at the camera. It's got a tiny processor and limited power and cooling. Encoding from a 21mp sensor(effectively 7mp) to 2k resolution is unlikely going to be stellar video. It's great looking not for the prowess of compression and resolution but the DOF and lenses available. Let's not look beyond what it is capable of. It's got excellent read noise control and would preserve detail where most other HD cameras would start breaking down. Give it a lot of light and the picture aliasing becomes more visible, more so to the post guys than the internet viewer.

The next version of FCP is mentioned for 'realtime editing' of H.264 however

That's only with AVC-I 50/100. Not with AVCHD and other H.264 formats like the 5D2 variant which still require transcoding. I think it'll be worthwhile to consider the FX series just for the faster transcoding with Elemental Accelerator.

[Steve Douglas]

Hey Jim,
The dolphin footage looked fantastic. Yes, Pro Res is easier to work with despite the larger file sizes. With the new Final Cut Studio now having Pro Res 4444, you might find it even easier and being able to maintain quality.
Steve
www.kenstone.net