jcs

I have two 5D3's; my primary goal is fast production and minimum file size and storage, with minimum time in post (shoot close to desired look). Many times I also shoot stills when shooting video on location. My comments in bold, below. What you gain by choosing the Blackmagic: At which point I need higher quality than the 5D3, I'll be looking at renting a C300 (or purchasing if it makes sense for the project(s)), Sony FS700 or similar, latest Panasonic large sensor, etc. That is, high quality with an excellent long GOP codec. For something bigger budget, Arri Alexa or F65 (possibly Red if reliable); perhaps BMC in that case as B-cam(s). My requirements won't be the same as everyone else's. Different cameras for different jobs... [*]Stills [*]Wide angle faster than F2.8 [*]Full frame rendering of your lenses [*]Usable ISO 12,800 [*]Very small files which can be edited in real-time. [*]No need to transcode. [*]Fantastic audio when using a high-quality preamp. [*]Excellent handling for solo- handheld, and run & gun. [*]Incognito video recording. [*]420 recording sufficient for material destined for internet, TV, and Blu-ray. [*]Relatively low noise until very high ISO's: remaining noise cleans up well with Neat Video. Shooting a band in near darkness at F4 is possible with ISO 12800 and Neat Video. Faster lenses allow even more light (while narrowing DOF, sometimes desired). [*]Raw codec: massive files and a great deal of cumulative time to process. [*]Double the resolution (1000+ lines vs 600 lines): so far no chart posted for BMC. I read the posted charts at 800+ line pairs before extinction for the 5D3. I will interpret the BMC using the same criteria once an ISO 12233 chart is posted (obscure German blog link posted in this site did not include a chart, nor could I find "600" written anywhere on the page via search). [*]13 stop dynamic range: valid point. [*]12 bit colour: valid point, though more tests needed to see how much of those bits carry useful information vs. noise. [*]4-2-2 sampling: helpful for theatrical releases, not as helpful for material destined for 420 such as the internet, Blu-ray, and TV. Might help with green/bluescreen, however 5D3 material keys fine in my tests. [*]HD-SDI: helpful for pro monitoring or external recording (not an advantage for my needs). [*]Robust 1/4 audio jacks: we need to do tests to determine the quality of the preamp. 5D3 preamp at +1 gain works great with external preamp (many of which are small and can be mounted on the camera). 5D3 also includes headphone monitoring (available on BMC?). [*]Da Vinci Resolve (it comes in the box): excellent software worth ~$1000 by itself. For my current needs, too much work vs. editing everything in real-time from within Premiere CS6. A "Resolve" plugin for CS6 would be interesting. [*]Significantly larger screen and easier focussing: valid point, however I have gotten pretty good taking advantage of my nearsightedness using the 5D3 screen without an eyepiece. [*]Peaking and one touch focus assist: valid points (5D3 will get peaking with ML at some point (I don't use ML, though)). [*]ProRes recording option: will reduce file footprint however ProRes files are still massive. I prefer efficient long GOP codecs: modern quality is fantastic. C300 50Mbit/s quality is stellar (some folks claim long GOP is detectable by eye vs. ALL-I- let's do a controlled study :). All content on Blu-ray, TV, and internet is long GOP). More likely user's computer(s) too slow to decode long GOP fast enough, causing stuttering.

AFAIK- [color=#282828][font=helvetica, arial, sans-serif]H.264 4-4-4 10bit long GOP isn't supported yet- perhaps someday soon.[/font][/color] [color=#282828][font=helvetica, arial, sans-serif]I'd believe 600 lines at ISO 12800 if you could show it with a line chart. The example I posted is sharp enough given the conditions.[/font][/color] [color=#282828][font=helvetica, arial, sans-serif]Again the 5D3 isn't skipping pixels, else we'd see aliasing. Soft due to binning (averaging, but not skipping pixels) is likely. Averaging pixels is low pass filtering (cuts high frequencies: detail).[/font][/color] [color=#282828][font=helvetica, arial, sans-serif]Downsampling a still frame and comparing to a video still is something I have been recommending for a long time. It shows what is possible with the sensor, given enough processing power. I agree that cameras advertised as 1080p should actually provide 1080p, or 1000+ "TV Lines" (horizontal line pairs). There has been confusion regarding 1080p and specs of 1000+ TV Lines: 1080p refers to vertical resolution and and 1000+ TV Lines refers to horizontal resolution. To honestly capture 1920 pixels, we need at least 960 TV Lines (horizontal line pairs).[/font][/color] Here again is Jason's ISO 12233 line chart test: [url="https://vimeo.com/39517721"]https://vimeo.com/39517721[/url]. Post sharpening is fair to remove the effects of the strong AA filter. If someone posted a line chart for the BMC it would show the relative performance differences between the cameras in terms of resolution, acuity, and aliasing. My $700 (when new a few years ago) Panasonic TM700 is at least 100 lines more resolution than the 5D3 (~900 line pairs). It does great in bright light, but degrades rapidly as light diminishes. ALL-I: [img]http://www.eoshd.com/comments/uploads/inline/16978/50370de5de8e7_ALLIsm.jpg[/img] IPB: [img]http://www.eoshd.com/comments/uploads/inline/16978/50370e0a9cd38_IPBsm.jpg[/img] Full sized images here (mid page): [url="http://cinema5d.com/viewtopic.php?f=30&t=39999"]http://cinema5d.com/...hp?f=30&t=39999[/url]

[quote name='EOSHD' timestamp='1345760598' post='16337'] - 5D3 is 720p at best. It is nowhere near a 1080p camera let alone 2.5k [/quote] How do you know this? The 5D2 was rated around 700 lines (from a line chart), that's horizontal lines pairs, equaling 1400 pixels. Thus, 1400x788 resolution. The 5D3, from my and Jason's chart tests is at least 800 lines, to my eye it's 850, or 1700x956 resolution. Folks will argue about first aliasing artifacts and the detail extinction point- that's why I'm asking for charts for all cameras, so we can make unbiased relative comparisons between cameras (where each person can define their own first artifacts and extinction points). [quote name='EOSHD' timestamp='1345760598' post='16337'] - The codec has tons of mosquito noise in the lows and mids [/quote] As shown in multiple tests, the ALL-I codec which you prefer is indeed (sometimes) problematic for high detail, low motion scenes. The IPB codec does a much better job. A firmware update could improve both codecs. More info along with images: [url="http://www.dvxuser.com/V6/showthread.php?278746-ALL-I-vs-IPB-IPB-Wins"]http://www.dvxuser.c...vs-IPB-IPB-Wins[/url] [quote name='EOSHD' timestamp='1345760598' post='16337'] - The sensor still skips too much data [/quote] If the sensor skipped too much data, we'd have aliasing. The 5D3 is one of the best cameras to prevent aliasing. I would agree that the image is perhaps over filtered (smoothed) to protect upline cameras, or perhaps binning (averaging, low pass filtering) causes the same effect (Canon has not posted a whitepaper on how the video is processed). Post sharpening works great to undo the strong AA. If the 5D2 is good enough for the internet, TV, and the big screen, the 5D3 is even better. What does this have to do with the BMC? Nothing- responding to your comments about the 5D3. The BMC is another tool with its own pros & cons. [quote name='EOSHD' timestamp='1345760598' post='16337'] - The noise grain is blotchy [/quote] Here's ISO 12800, shot in almost complete darkness at the House of Blues, Sunset Strip: [media]http://www.youtube.com/watch?v=TbzF2e5beS4[/media] processed to be much brighter than what the eye sees. It was a bit noisy: Neat Video works great. Where's the blotchy noise? What does this have to do with the BMC? It's something it can't do- extreme low light (the sparkler example is very bright in comparison). Did you add noise or is the noise in the bright example BMC images from the camera? A clean image allows one to add noise grain of choice in post when desired. Reasonably sharp, detailed, and alias free even after youtube compression? [media]http://www.youtube.com/watch?v=qkFmhKoUEv4[/media]. Vimeo- with original MP4 download: [url="https://vimeo.com/39124220"]https://vimeo.com/39124220[/url] Each camera has strengths and weaknesses. The BMC will certainly fill a need. One of its greatest strengths for some is a deal breaker for me: raw and associated file size, processing, and data management. I would much prefer H.264 444 10-16 bit (long GOP). The spec supports it. Even H.264 422 10-bit would be great. Nature favors efficiency. The quality from 50Mbps long GOP 422 8-bit is amazing. H.264 efficiency recently doubled: [url="http://phys.org/news/2012-08-mpeg-codec-halves-bit.html"]http://phys.org/news...halves-bit.html[/url].

Andrew- I'm focused on a single factor: resolution (also relates to aliasing performance). Not discussing cost (pointed out earlier) or color depth. Per the math and in theory, the 3.8K C300 has enough samples to provide full 1920 resolution (960 lines) without aliasing. For the 2.5K BMC to outperform the C300 would mean something is broken in the C300 implementation (from examples I have seen, the C300 appears to provide more resolution than the BMC). Post-sharpened, the 5D3 is decent, and the BMC may very well provide more resolution. However, the current images posted are noisy; hard to determine actual resolution from close crops of faces. Wide shots with lots of detail are more challenging test cases. My opinion doesn't matter nor does your opinion matter: everyone perceives things differently. Math, logic, and the Scientific Method were created as a means to allow people to reach agreement, to discover new things, etc. From Zen philosophy we can throw out math and logic, however agreement becomes difficult unless all parties are already on the same wavelength. Test charts provide a means for which we can reach consensus, especially when we can compare charts from multiple cameras side-by-side. While we may disagree on maximum resolution as it approaches extinction of details, we can reach agreement on relative performance.

When someone makes statements that a device is high resolution, including higher than not just the 5D3 but also the C300 (which has a 3.8K sensor (the amount required for full 1920 sampling without aliasing) vs. the 2.5K BMC, a logical request is for a resolution chart for an non-subjective comparison. I own Sony, Canon, and Panasonic cameras (include a BMD Intensity card). I use what works- they are all tools. Shooting with similar sharpness lenses side by side is also a good test (but will tend to be more subjective). Why would anyone not want to see the truth? A resolution chart for cameras is like an MTF chart for lenses: both very useful metrics in understanding hardware performance. Subjective tests are also valid, but completely different tests. I'm asking for a simple ISO 12233 test chart shot with the BMC to compare with other cameras (GH2, 5D3, F3, FSx00, C300, etc.). Here is a much more detailed independent analysis of the C300: [url="http://blog.creativevideo.co.uk/wp-content/uploads/2012/01/canon_c300_alan_roberts.pdf"]http://blog.creativevideo.co.uk/wp-content/uploads/2012/01/canon_c300_alan_roberts.pdf[/url]. Based on the hardware design and test performance, it's not likely that the BMC can match the C300 in resolution and aliasing performance (not comparing cost, only responding to a statement that the BMC produces higher resolution images vs. the C300).

Andrew posted that this device has excellent resolution- a chart will show us the actual resolution: this is truth, and will show us how well 2.5K Bayer is UPSAMPLED to 1920x1080 (not enough photosites to downsample). The request for a chart is based on logic; many of the comments regarding not wanting to see a chart are based on emotion. This is a technical analysis blog post and forum thread. Emotion makes most sense when discussing content (as opposed to technical specs).

No need for a subjective debate, here's a reasonable 5D Mark III resolution chart test for an empirical comparison*. Look forward to seeing what the BMC can do. Resolution is very important to many folks- a chart will show us a good deal of info regarding resolution, acuity, and aliasing performance. Resolution has been huge source of debate on this site. Recall the images I processed showing the non-OLPF removed footage closely matching OLPF-removed footage before it was decided OLPF-removal wasn't helpful. I also showed sharpened 5D3 footage from Tonis' early 5D3 footage was decent, before sharpening was considered an acceptable option (for many folks it wasn't and still isn't). Therefore, it is reasonable to request a resolution chart test for the BMC when it is claimed it is much high resolution than the 5D3, which has been heavily bashed on this site for poor resolution. 2.5K native Bayer isn't enough for 1920x1080: we need to sample at least 2x, or 3840x2160 per photosite (RGB), per Nyquist sampling theory. Again, a resolution chart will show us how far they got via Bayer interpolation. [url="https://vimeo.com/39517721"]https://vimeo.com/39517721[/url] * folks can decide for themselves the method of extinction of detail, using the same metric for both devices.

Yes- how about a resolution chart shot? Then we can compare apples to apples.

For grading looks like a nice tool. Compared to footage I have shot and edited (+ post sharpened) with a 5D3, it doesn't look any sharper or more detailed. Any way to compare at this time?

The anisotropic antialiasing of the 1DX makes the image look 'funny'- like something is wrong with the lens or image processing: not natural. The 5D3 footage, while softer, looks natural (and sharpens well in post).

It's possible to create USM (and LCE) in PPro which is 100% GPU accelerated and runs in real-time: 1. Create a copy of the clip and stack above the original. 2. Apply a Gaussian Blur to original. Set to 10 to start. 3. On the upper clip, apply a Brightness & Contrast and set Brightness to 63.7 and Contrast to 50. Set Blend Mode to Difference. Adjust Gaussian Blur on original to control radius, adjust Opacity on the top video to control amount. It's not quite the same as the built-in USM, however it shows how it's done and that Adobe could include a GPU accelerated version instead of the slow CPU version. Also try adding an Adjustment Layer (CS6) or Nesting (CS5.x), and applying a (convolution) Sharpen to the Layer or Nested clip. Tested this technique in CS6 with a 1280x720@60 clip and it brings detail up significantly (and runs in real-time): can easily cut with 1080p material. It's too bad PixelBender effects can't run in PPro (must create in AE then use Dynamic Link, which is slow). With general GPU-shader support, PPro would be a much better package, providing a means for much more real-time GPU effects. Virtualdub+avisynth et all are powerful tools and run pretty fast, but require round-tripping footage. As a time saver, I'm limiting my effects/tools palette to what runs in real-time directly in PPro (Neat Video & Warp Stabilize in CS6 are the only two exceptions).

[quote author=Andrew Reid link=topic=913.msg6660#msg6660 date=1340969767] Interesting that you found less macro blocking in IPB mode JCS. What about transcoding ALL-I to ProRes with 5DToRGB? Does that solve the decoder issues? [/quote] Transcoding ALL-I to ProRes might have brought ALL-I up to IPB quality in terms of macroblocking (tested in OSX; reported results on other forums), but it did not surpass IPB. Additionally, ALL-I was noisier. Noise steals bits better spent on detail. CS6 appears to do a better job with macroblocking and ALL-I. However A/B compared 1080p ALL-I to IPB could not see an improvement with ALL-I (low motion shots). 720p60 ALL-I was clearly better than IPB, however. It's clear that long GOP codecs can be very efficient and very high quality for both static and motion shots: the XF camcorders and C300's 50Mbps MPEG2(!) 422(!) codec performs extremely well. That would be an awesome codec for the 5D3. Perhaps biz/marketing reasons won't allow it. 422 H.264 theoretically could be even better than MPEG2, however while supported by the standard, has not been implemented yet (would have compatibility issues).

Thanks Axel. Monty Python's [i]The Meaning of Life[/i] is classically irreverent, as with all of their movies ([i]Holy Grai[/i]l & [i]Life of Brian[/i] are my favorites). They make fun of the question, as they should. Douglas Adams' [i]The Hitchhiker's Guide[/i] series wonderfully brings out the absurdity of life in civilization. [i]The Tree of Life[/i] and [i]The Secret[/i] did not directly influence the script (heard [i]The Secret[/i] audio years ago, didn't see [i]The Tree of Life[/i] until after I wrote the script). After showing initial cuts of the movie to people, some recommended watching [i]Zeitgeist[/i] & [i]Thrive[/i]. I can see why folks make the reference, though [i]How to Live[/i] aims to simplify complex ideas into the simplest, easiest to remember forms possible. Many religions and ancient philosophies hint at these ideas, but are much more complex and without the root concepts in math, science and iterative computation. That is, we can point at the universe and have common agreement regarding what the universe is doing: we can each observe, measure, and compare to what others said, and find agreement in perception. That is the basis of math and the scientific method. However, as stated in the beginning of [i]How to Live[/i], math (and science) cannot fully describe the universe. Problems exist beyond human logic and reasoning (proven by Gödel, who went insane). Zen philosophers figured this out over 1500 years ago: certain forms of knowledge can't even be put into words or language, let alone mathematics. In a heavily rule-based society, Zen provides a means of finding peace in a society that claims to be fair, just, ordered, and every person is treated equally, when in fact it is really unfair, highly chaotic, and people are not really equal. I found Zen very helpful when working for Newscorp/Fox/Myspace- the largest corporation I have worked in. When studying Zen, Zen masters appear to say and do silly things to/with their students to help them break old patterns of thought. Thus it makes sense what Monty Python, Douglas Adams and other have to say about living in this universe. The title of [i]How to Live[/i] was also inspired by Zen: it was the simplest, shortest title possible. The next episode goes into more detail of the nature of the universe. Later episodes and interviews will discuss mind/body/nutrition. All of these concepts will relate back to the simple, two generative concepts of the first episode.

We finished our first project with the 5D3. Tagline: "What is the meaning of life?" Perhaps we should ask a different question. http://youtu.be/BLbZ_nwm8-I [b]Vimeo version with full HD download[/b]: [url=https://vimeo.com/44909002]https://vimeo.com/44909002[/url] Lenses: 16-35 F2.8 II, 24-105 F4L, 70-200 F2.8L II. FaderND gen 1. Profiles: Faithful and CineStyle. VO: Audio Technica 4029, Mogami cable, FocusRite Scarlett 2i2 (very low cost and ultra quiet!) into Reaper on OSX. Edited in PPro CS6 on Win7x64, image stabilized and post sharpened.

There are plugins which claim to fix clipping- google can show them to you: here's one: http://www.izotope.com/products/audio/rx/ (haven't used it).

Here's the best grass I could do with the 5D3: https://vimeo.com/40200361. I believe the circular polarizer in the Gen (1!) Fader ND helped with improved accutance (microcontrast). The 24-105 F4L is not ultra sharp, yet the grass after post sharpening is just starting to alias. The 1DX has weaker horizontal antialiasing (vs. vertical as noted in the article). In computer graphics we call this 'anisotropic' filtering (http://en.wikipedia.org/wiki/Anisotropic_filtering), which means the filter is not the same in all directions. If one desires to post sharpen 1DX footage, performing a horizontal blur (Guassian, etc.) before sharpening can help prevent aliasing. For relatively low-motion 1080p shots, IPB produces less artifacts than ALL-I (except when shooting 720p60: ALL-I is better). I've seen ALL-I break 101Mbps (600x Lexar Pro CF), but it still can't match IPB quality for low motion shots.

[quote author=Andrew Reid link=topic=884.msg6451#msg6451 date=1340552541] Good spot, it certainly does seem like there's a change to the anti-aliasing filter(s). From what I understood about the 5D Mark III it has a two-part filter, one to blur the horizontal and one vertical. However I'd say the gain here is due to reading more pixels, rather than anything in the optical path. [/quote] It would be helpful if Canon released whitepapers for the 5D3 and 1DX, as with the C300, explaining how optical and image processing is performed. Anisotropic filtering for texture maps on an angled, 3D projected image is a useful feature, however it's not desired for screen-aligned 2D images. After a small amount of post-sharpening on the 1DX, much of the horizontal antialiasing is gone. Sharpening on the vertical axis only will preserve the horizontal antialiasing. Another option would be to perform a horizontal blur first, followed by a normal sharpen. Extra work in post... A challenge which can be directly compared to other cameras: how does it do with the ISO 12233 chart compared to a downsampled still frame? Since many "1080p" (1920 horizontal res) cameras aren't storing 960 horizontal lines (minimum required for true 1080p), it would be helpful if resolution charts results were collected in one place, along with commentary (as we do with audio gear: http://www.avisoft.com/recordertests.htm). Once the list is constructed, a leaderboard is established, and provided the page is popular and well-known, manufacturers will strive to be on the top of the list.

Examining the full clip in detail, the 1DX has more resolution, however its antialiasing is weaker. It appears to be antialiased more on the vertical axis than the horizontal axis, giving it a more digital look vs. film. Ideally, the image is high resolution and antialiased. The 5D3 image looks more like film, even though it is lower resolution. Both images were noisy, and the ALL-I codec did not work very well for this shot (low motion): IPB would likely do better in this case with more detail and less block artifacts. 5D3 image sharpened at 33 + minor curve adjust, 1DX sharpened at 16, both Neat Video noise reduced. After processing, during playback and full screen, hard to see much of a difference in quality.

Don't have time for extensive tests, however in CS6 720p60 looks a bit better using ALL-I (detail). 1080p24 looks the same or IPB might be better for slow moving shots (detail). For CS6, it's no longer necessary to rename MOV to MPG to get real-time performance.

[quote author=peppebadala link=topic=621.msg5202#msg5202 date=1336700183] resolution does improve... i still dont know why people are saying it doesnt... [/quote] Can you test with this chart: http://www.graphics.cornell.edu/~westin/misc/res-chart.html and post the results? See also https://vimeo.com/39517721

[quote author=Andrew Reid link=topic=621.msg5276#msg5276 date=1336882145] This video has been deleted. Anyone know why? And does anyone have a copy? [/quote] I have a copy (63MB). The resolution and quality was worse than stock. Can you test your no-OLPF (single) 5D3 on the same resolution chart (http://www.graphics.cornell.edu/~westin/misc/res-chart.html) and post the results? One thing that may be happening is while resolution may not increase, acutance might.

For simple edits, 1GB GPU memory is fine. For 4K and complex edits, >= 2GB is needed. My 1GB GTX285 could not render a recent project, however a 2GB GTX285 could with some issues and work-arounds, and a 2.5GB Quadro 5000 had no issues.

I tried both SD (95MB/s which is capped at 20MB/s ave. on the 5D3) & CF (Lexar Pro 600X; ~45MB/s ave.); quality was the same. No errors with SD. However I don't use ALL-I anymore after initial testing due to poor quality compared to IPB in CS5.5. Just installed CS6, will test ALL-I again and see if decoding quality has improved.

I found 60fps film to look excellent: Showscan (by Douglas Trumbull: http://en.wikipedia.org/wiki/Showscan). While it did not work out for traditional movies, they did make it into a successful business for theme parks and location-based entertainment: http://www.showscan.com/ If the Hobbit doesn't work at 48fps for whatever reason, a trivial drop-frame to 24fps will fix it. Stereoscopic panning definitely works better at higher frame rates, though. However, there is an alternative experiment to try: variable frame rate in the same picture. Show outdoor shots at 48, and other shots which don't work at 48 at 24. One thing we have discovered is that if we bring a simulation (or movie) too close to reality, but not quite there, it has a strong negative reaction, psychologically: http://en.wikipedia.org/wiki/Uncanny_valley. It may be that we'll get used to it, like beer (remember the first time you tasted it?). Certainly for outdoor nature shots, high resolution, high framerate material makes the front of the theater look like a giant window into the real world. Add some fans and fragrance and the experiences becomes more real. That's where Showscan ended up (including motion platforms).

Bill DePalma removed both OLPFs and had the glass replaced and dust system replicated (via Maxmax.com): https://vimeo.com/40789276 Andrew- you removed only one OLPF and did not add replacement glass (nor re-attached the dust system)? Still curious if the software changes behavior with the dust system connector removed.