jcs

[quote author=Sara link=topic=611.msg4406#msg4406 date=1334891961] Opinions will differ on skin tone I guess, but to me those are good for a Canon 5D but still waxy / videoish / and very cheap looking.  I watched them on an Eizo ColorEdge CG245W. [/quote] We can remove opinion and subjective evaluation for skin quality by first understanding how to create skin with math (computer graphics): http://http.developer.nvidia.com/GPUGems3/gpugems3_ch14.html. Once we have reached agreement on what looks accurate/realistic, we can discuss how to transform away from realism into video (limited dynamic range, bunched/crushed highlights, limited pixel detail (blurred to remove noise & detail, then sharpened with unsharp mask), edge halos/bloom/glow), aliasing, etc.) into a film look (extended dynamic range (HDR), no crushed highlights, detailed pixels, no halos, zero aliasing). The BMD examples have aliasing and I'm not yet seeing HDR, or HDR effectively tone-mapped into 8-bit for the web (there is color banding/solarization below the left eye and on the neck/shoulder). There are frames in the 5D3 Copelandia footage where limited DR and compressed highlights hurt the look, however there are plenty of frames (the eye close up, for example) which look excellent. Some shots of the girl's face (Christopher's test footage) have highlight peaks and blooms/halos created by in-camera sharpening (unsharp masking), which hurts fine detail (which he acknowledged in the comments). How do we simulate plastic in CG- we make sure there is no sub-surface scattering, don't render fine detail, and use a simple Phong model for specular highlights. Rendering realistic skin requires modeling sub-surface scattering and a sophisticated energy-accurate specular model. Note that the BMD model example is wearing significant makeup- not a rendering of skin but rendering make up on skin (oil and sub-surface scattering are hidden). The hands, with no make up have a video look: low detail, bloomed specular in thumb creases (even though not blown out completely). Looking at the cuticles, there appears to be a biotin (B7) and B12 deficiency.

The colors look fine, however in another thread with the same image I observed block artifacts on the hands. Folks are stating this camera is a game changer; perhaps best to compare with other cameras on the same material at the same time (same conditions). Skin tones look pretty good for one of the 'crushee's', the 5D3: https://vimeo.com/39237997 https://vimeo.com/40117480 I too would prefer more detail for for the 5D3, however for up-close, skin tones, etc., it works pretty well.

[quote author=cameraboy link=topic=596.msg4188#msg4188 date=1334694708] i expect excellent debayer option  in resolve .... [/quote] A resolution chart will let us know...

[quote author=cameraboy link=topic=596.msg4120#msg4120 date=1334654481] after debayering RED  4K IS about 2.4k so we can expect 1.6K from BMD camera (not bad at all) http://provideocoalition.com/index.php/awilt/story/more_red_res_testing_the_mysteryium_resolved/ and all raw files look soft even RED EPIC (u need to add little sharpness) and it looks like that they use Fairchild sCMOS Sensor(its only sensor that fits BMD camera size) ...wow its rated at 88db of dynamic range ...thats more than ALEXA... http://www.scmos.com/ [/quote] That's a cool test chart showing RED 4K using decent de-Bayering to get to 3.2K before detail extinction (IMO, easiest way to mark end of useful resolution. First artifacts perhaps start at 2.4K, but there's a lot more usable information beyond that). This is to be expected for a B&W image (luma), however chroma resolution is lower  (and shows quite a bit of artifacting, especially with NR turned off). Thus, 4k/3.2k = 1.25. If BMD has an ultra-high quality de-Bayer algorithm (perhaps better than VNG (Variable Number of Gradients)), which can take a bit of time to process (though if can be implemented on a GPU (BMD is strong here (Resolve)), then 2.5k/1.25 = 2k, which would be perfect for 1920x1080. Would be great if BMD published a horizontal lines (resolution) spec (C300 and FS100 are rated at 1000+ lines (horizontal resolution)). 3840 using SuperPixel de-Bayering is best for HD (and also very fast), however 2.5k VNG or similar isn't a bad bang-for-the-buck. A resolution chart will give us a better idea for real-world performance.

The reason the video is soft is 2432 Bayer photosites means about 2432/2 = 1216 de-Bayered pixels vs 1920 from 3840 Bayer photosites (C300 et al). When companies advertise 4k, and the sensor is a Bayer array, you're only going to get around 1/2 the resolution (might be slightly higher with software interpolation tricks (e.g. for luma), but fundamentally it's about 1/2 resolution). If binning is being used to increase dynamic range and low light performance, resolution will be even lower. http://magazine.creativecow.net/article/the-truth-about-2k-4k-the-future-of-pixels Thus, if you want a true 1920x1080p camera, you need at least 3840x2160 Bayer photosites (as with the C300). This camera[s]might[/s] will have more dynamic range vs. a 5D3 (due to 12-bit RAW and 10-bit ProRes/DNxHD support), but it won't be any sharper (looks softer from the example footage so far; makes sense given the sensor resolution). "Downfall of the big guys"- not yet. This will happen when folks start making modular camera systems: lens mount, sensor, processor, and storage, as separate components. That's when things will get interesting, just like  when the IBM PC came out and the clone wars started. Who's going to make the best camera OS? Companies like NVIDIA and AMD will start adding GPUs to cameras, etc. This can be done somewhat today with a laptop and an industrial camera sensor over USB3 (and later Thunderbolt). Some early work in progress: http://graphics.stanford.edu/projects/camera-2.0/

If there is any real effect, I think it involves the polarizer. Testing a linear polarizer (in addition to circular), at different angles would also be interesting. Check this out: http://nikonusa.com/en_US/IMG/Images/Learn-And-Explore/2012/Camera-Technology/D-SLR-Series/Moire-D800-D800E/Media/OLPF_schematic.pdf A polarizer might stop/reduce/change the last diffusion stage (if the 5D3 works similarly). A collimator might also have interesting properties: http://en.wikipedia.org/wiki/Collimator (check out the Söller collimator drawing): [img]http://upload.wikimedia.org/wikipedia/commons/5/52/Collimator.jpg[/img] Ah, interesting, another clue from the Nikon link: [QUOTE]By converting polarized light into circularly polarized light with the wave plate, two points are divided into four points at low-pass filter 2.[/QUOTE] Since a circular polarizer is a linear polarizer with a 1/4 wave retarder (phase shifter, "wave plate"), curious if this might explain a real effect when a circular polarizer is placed in front of this optical assembly. Intuitively, based on testing, it could be a circular polarizer reduces the effect of the OLPF by 1/2 to 1/4. If the OLPF works by providing two light paths, perhaps pre-polarizing produces one light path (or narrows the net separation).

[quote author=Andrew Reid - EOSHD link=topic=561.msg3701#msg3701 date=1334136226] It is possible the electronic contacts to the OLPF trigger a stronger anti-aliasing blur when video mode is active. More likely, video is more effected because you are viewing the results at 1:1 on a 1080p screen. If you look at stills they are quite soft at 1:1 as well. [/quote] This can be tested by pulling the connector and leaving the OLPF in place.

[quote author=Andrew Reid - EOSHD link=topic=558.msg3680#msg3680 date=1334105485] I'm using a SDHC 95Mbps and get 66-88Mbit footage in ALL-I mode. What are you using to analyse and playback the footage? It must surely have a problem with ALL-I because there is no WAY the highly compressed IPB has better image quality from my experience. [/quote] I've gotten >101Mbps using 600X CF Lexar Pro with ALL-I (IPB had less artifacts (looked better) at 30Mbps, same scene & conditions). Regarding IPB- the tree/sign scene below shows IPB outperforming ALL-I. The 400% crops posted here (see attachments) and elsewhere show macroblocking issues with ALL-I but not IPB. Some folks can see a quality improvement with IPB even on the youtube footage. The best I have seen with ALL-I was using Compressor (old version from FCP7) to transcode before analysis. It came closer to IPB but still had a few hard macroblock edges (same scene as below). If you have shot and analyzed ALL-I compared to IPB, and found IPB inferior, can you post crops to show ALL-I outperforming IPB along with the settings used? I have tried low- detail, high-detail, motion, still, etc., can't get ALL-I to perform better than IPB (using Faithful or CineStyle).

Just did a non-scientific line chart test. It's very subtle (not worth posting pics), but it looks like slightly less fuzz on the lines near the limit when using the polarizer (variable ND). Since the outer polarizer can be dialed, lots of variables (could test on an HDTV live and perhaps check for sharpness changes). Also tested ALL-I- looked terrible compared to IPB: macroblocked everywhere (to be fair, was using SDHC (95Mbps card limited to 20MBps by 5D3) and not CF (can hit 45+MBps), though the ALL-I clip hit 48Mbps (6MBps). Again, sharpening in post requires about 1/2 normal with bright outdoor shots with the polarizer in place. Regarding IPB- if you can't see that it's better at 100% crop, then ALL-I has no advantage (except for slow computers). However, when you study IPB and ALL-I at 200-400% crop (and later start to see macroblocking in ALL-I at 100% once you know what to look for), ALL-I becomes even less attractive.

[quote author=moebius22 link=topic=561.msg3673#msg3673 date=1334103053] Wait till somebody figures out they can do this for money. [/quote] $450 USD + shipping (includes optical replacement glass): http://maxmax.com/IRCameraConversions.htm

Compare the grass up close (ALL-I and IPB) to this footage, same lens, 24mm: http://www.youtube.com/watch?v=eTEV6_os4gY Normally grass is mush- looks less mushy with the polarizers (ND affect achieved by two polarizers out of phase (either a linear and circular (made from a linear polarizer and 1/4 wave retarder), or two circulars)). Again, I shot more than this clip, and each clip with the polarizer needs less sharpening. Lots of folks have polarizers & variable NDs- should be easy to validate or invalidate by testing.

This scene shot at 24mm with a FaderND gen 1 (works fine with the 24-105 lens) looks sharper than normal for a detailed wide angle shot: https://vimeo.com/40107424 It's not sharp, but it appears sharper than normal and requires less post sharpening (about 1/2 the amount). This example shows moire (church), which I haven't seen before with the 5D3 (well, have seen it before but very minor). More details in the vimeo description and original MOV available for download. Banding in sky likely due to using CineStyle and over-exposing the shot (could fix in post if this shot was needed using noise/grain or selective blurring with AE or Resolve). If anyone has a FaderND (any version), Heliopan, etc., or clear linear/circular polarizers, would helpful to know if polarizers can help improve sharpness (some OLPFs use a polarizer- might be some kind of interaction with polarized light which reduces diffusion and blur).

[quote author=brice link=topic=501.msg3307#msg3307 date=1333240640] Sorry for being harsh, just not my cup of tea. But I am wondering what all was done to the screenshot? Given that the text was embedded in the file you worked with, I am wondering why the text looks like it has a slight Gaussian Blur applied? [/quote] Hey brice, no worries. It was sharpened in PPro CS5.5 using the "Sharpen" filter (convolution), some RGB curves, then an organic GPU accelerated film grain was added (real-time): [url=http://www.dvxuser.com/V6/showthread.php?278876-Film-Grain-Real-time-GPU-PPro-CS5-5.]http://www.dvxuser.com/V6/showthread.php?278876-Film-Grain-Real-time-GPU-PPro-CS5-5.[/url] The text was originally slightly blurry, but got sharp after sharpening, blurred it in Photoshop before saving to JPG for the web. You kind of raised a good point: I should have just matched the WITHOUT-OLPF image sharpness.

[quote author=brice link=topic=501.msg3301#msg3301 date=1333234858]Wait, what? Much better? Sorry, but this looks horrible.[/quote] That's cool. All of this is art. I like it (I oversharpened it to make a point: could have been matched to the other image for a more pleasing film look). What part(s) don't you like? The lighting and color between the images was different from the start. I processed it to show high frequencies present in the WITH OLPF dataset (image frame). It's not clear yet if removing the OLPF will result in any more information being stored after processing and compression. A resolution chart would show with reasonable certainty if resolution has been improved past ~890 lines (measured from multiple sources with resolution charts).

[quote author=Andrew Reid - EOSHD link=topic=501.msg3293#msg3293 date=1333225032] What have you done to that frame grab to make it look so bad?[/quote] You're right, that version doesn't look very good: tried to match brightness. Sharpening was in Photoshop: USM. This version is much better, done in PPro CS5.5 using convolution sharpen and my GPU noise effect. To my eyes, the processed OLPF-present image looks pretty solid. There are regions which look better in the OLPF-removed image (front face brick areas, etc.), however I'm not sure those aren't related to using ALL-I (was James using ALL-I or IPB)?

Andrew- can you guys shoot a resolution chart? Curious how no OLPF and no post sharpening compares to OLPF and post sharpening (resolves about 890 lines (horizontal resolution) post sharpened). Also, I see lots of ALL-I looking compression artifacts (macroblock quantization). Can you try IPB (my tests show stores more information, less noise, and less artifacts (+ ~3x less diskspace). Finally, you can get much better post sharpening results by turning off in camera sharpen: this footage has halos. Note that even with the compression artifacts and lower light/contrast on the OLPF image, it looks excellent post-sharpened compared to no OLPF (not clear if its worth removing vs. post sharpening. A line chart test would be helpful).

Hey Andrew- I did further deep analysis due to this thread: [url=http://cinema5d.com/viewtopic.php?f=30&t=39999]http://cinema5d.com/viewtopic.php?f=30&t=39999[/url] Log in to see clear pictures showing significant problems with ALL-I vs. IPB (pics also attached below: ALL-I is first). ALL-I has significant noise issues, and the bigger problem is excessive high-detail loss due to a macroblock quantization issue (even when the camera is putting out 80+Mbps; ~30Mbps IPB is significantly better (see 400% crops)). IPB looks beautiful and uses ~3x less disk space. The quality reduction in ALL-I is even visible after youtube compression (watch fullscreen, 1080p): [url=http://www.youtube.com/watch?v=eTEV6_os4gY#ws]Canon 5D Mark III - I-Only compared to IPB[/url] (even more visible with original footage). The problem with in-camera sharpening is that it uses an Unsharp Mask technique, which creates halos that cannot be removed in post. There is no mathematical advantage and only a disadvantage to using sharpen in camera. With sharpening turned off, compression should be more efficient, and sharpening can be very extreme in post before artifacts appear (using either a convolution sharpen "Sharpen" in PPro, or Unsharp Mask with a radius of < 1.0). Thus my recommendation based on these tests and analysis is: [size=12pt][b]IPB, Faithful, 0,0,0,0, Manual White Balance[/b] Sharpen in post with "Sharpen" (convolution, real-time in PPro), or Unsharp Mask with radius < 1.0 (typically slower and not real-time). Larger values of Unsharp Mask can enhance mid frequencies, and really large values (50+) can create pleasing "Local Contrast Enhancement": [url=http://www.luminous-landscape.com/tutorials/contrast-enhancement.shtml.]http://www.luminous-landscape.com/tutorials/contrast-enhancement.shtml.[/url] [/size] Haven't tested any custom profiles yet (should work as expected).

Wasn't the GH2 measured at 800 lines or so? (resolvable horizontal resolution), and didn't you match the 5D3 to the GH2 with post sharpening? (removal of some of the 5D3's anti-aliasing) From my informal tests it looks closer to 800-900 lines of resolution (test chart) True, it's not 1000+ as rated by the top cameras, however it's far less expensive and is full frame (GH2 and FS100 aren't really the same class of device).

[quote author=Andrew Reid - EOSHD link=topic=456.msg2928#msg2928 date=1332626229] Thanks jcs. However I have decided I just don't like the digital sharpening. It should be 1080p straight off the bat, otherwise it looks weird. I am finding that I either apply a little sharpening, which hardly makes it much closer to true 1080p, or I apply a lot which works for some shots but completely destroys others. 5D Mark III is teetering on the bring of eBay for me right now since D800 is looking like a worthy replacement, even with the slight moire and aliasing it appears to have. [/quote] Hey Andrew- did you look at the Veyron video I posted? I sharpened as need per shot. Did you see anything weird or otherwise unacceptable? The D800 also appears to be a fine filmmaking tool (especially since it sounds like you have lenses you can use with it), however it will be interesting to see if Canon releases updated firmware to produce sharper, artifact-free footage straight out of camera (I can see a point in not wanting to sharpen in post if time is an issue, etc.). Until Canon tells us how they are processing the sensor data, we won't really know the true capability of the hardware. I'm also curious if there is any technical reason they couldn't use the 50Mbps 422 MXF codec (lots of business & marketing theories, though). I'd rather use that codec than 91Mbps I-frame (higher quality & less disk space). Looking again at your sharpened sheep frame, I see halos. When you say it looks weird, is this with footage shot with in-camera sharpening turned off?

hiphopsuperman- Yes- it's the "Sharpen" effect, in the Effects window (type 'sharpen' in the search box). Simco123- the built in sharpening appears to be a form of Unsharp Masking. It tends to work well in increasing perceived sharpness, however it can generate halos, which makes sharpening further in post more challenging. Turning off sharpening in camera gets you a relatively raw image, where you can apply a convolution sharpen ("Sharpen" in PPro) or Unsharp Mask, or both in post. Sharpen will enhance fine details, and if set too high will cause aliasing and can also make noise and other artifacts more apparent. Unsharp Mask works by subtracting low frequencies and won't have those issues, however at higher levels it can create halos. Unsharp Mask can also be used for Local Contrast Enhancement: [url=http://www.cambridgeincolour.com/tutorials/local-contrast-enhancement.htm]http://www.cambridgeincolour.com/tutorials/local-contrast-enhancement.htm[/url]

[url=http://youtu.be/qkFmhKoUEv4]http://youtu.be/qkFmhKoUEv4[/url] Bugatti Veyron grills and fine details, fine patterned clothing, brick walls/ground, wide angle but still sharp. Watch in 1080p full screen to prevent aliasing.

Canon 5D Mark III pointed at brick walls. 70-200 F2.8L II. "P" mode, "Faithful", IPB, 1080p24. Sharpened in PPro CS5.5 @ 33. Watch full screen 1080p. [url=http://www.youtube.com/watch?v=LWXim4E-z58#ws]Canon 5D Mark III pointed at bricks[/url]

It appears sharper out of the camera but has aliasing and moire- the 5D3 doesn't have aliasing and moire but looks softer. Makes sense. From your GH2 compare with post-sharpened 5D3 footage, they looked about the same sharpness, though the 5D3 had less noise and still looked more anti-aliased and filmic. Sharpening in post with PPro is basically free, and highly tweakable- don't see it as a problem. Some material will need extra help with aliasing and moire- if the camera can't handle it, the shot must be adjusted/wardrobe change etc. I dig high resolution, but prefer image quality first (little or no aliasing or moire). We can always remove anti-aliasing in post in exchange for more perceived resolution via a convolution sharpen filter. To get the filmic look, we must remove aliasing and moire, then we can add back in a pleasing noise grain to enhance perceived texture sharpness.

Looking more closely at the GH2 image: it's much noisier and highly aliased. The 5D3 footage is cleaner, and even after sharpening, looks well anti-aliased while being about as sharp as the GH2. I will shoot outdoor daylight footage during lunch today.

The two shots looks similar. The GH2 looks like a digital camera, the 5D3 looks more like film (also less noisy). Is that what you see? Turn down sharpening if motion has issues. Shot indoor with the 70-200 F2.8L II, camera setting P (as shipped), profile "Faithful" (everything 0), IBP. The footage in PPro looks excellent. Didn't really need sharpening (looks a lot like film), but added a Sharpen 33. File was renamed with a .MPG extension: plays in real-time. Note that the DOF is fairly narrow, however the small text is readable. Playing back full screen looks great. I ordered 2 of these cameras (kit and body only). Definitely keeping the kit. I need to shoot a project with it before opening the body. Goal is to be able to shoot with two cameras. Have not decided whether to keep the body yet (will know better after using the kit camera on a real shoot). Might replace body with a pro video camera (high quality image stabilizer, high quality autofocus, for interviews). Need to demo/rent them first; leaning toward something that shoots 50Mbps 422, XF series or similar (can't justify C300 yet).