jcs

Thanks for the tests as well as grading and sharpening the 5D3 footage- that's what most folks will do to get the most out of that camera. I should have an SB to test this evening with the FS700.

While I can significantly sharpen 5D3 footage in post (60+ "Sharpen" in PPro CS6), I can only sharpen FS700 footage around 10 or so (kit lens; waiting for Speed Booster to arrive), not that it needs it though. I think we all understand now that the 5D3 is really a 1620x910 or so camera vs. the FS100/FS700 being full 1920x1080 cameras. The 1DC (and 1DX) are slightly better in full frame, but still not full 1920x1080 cameras. I'm beginning to think perhaps it's a hardware design issue vs. business-feature-holdback. That said, the 5D3 looks great as a close up camera, especially with post sharpening (and will look a bit better when the 422 HDMI out firmware is released).   As dark and blurry as your example shows the 5D3, it looks more like film (analog) than the FS100 image (more digital artifacts, sharp transients- looks more like video). I'll try to match the FS700 + Speed Booster to the 5D3 look when I do further testing.   Attached is a quick tweak to the 5D3 shot- sharpened, brightened, noise reduced, color adjusted. Can't match a true 1920x1080 camera, but isn't as bad as the original. [attachment=389:5D3.jpg]

It's not just about full frame support (though I admit that's what got my attention as a 5D2/5D3 owner with Canon lenses). They named it "Speed Booster" and not "Full Framester": the light gathering and focusing nature of the device is very cool. My 50 F1.4 becomes an F1.0, the 24-105 F4L becomes an F2.8, 70-200 F2.8 II becomes an F2.0 etc. The last feature which most people find hard to believe is the increase in image quality (improved MTF- modulation transfer function). This is measurable (they show graphs), and example images with reduced CA, increased sharpness and contrast: http://www.metabones.com/images/metabones/Speed%20Booster%20White%20Paper.pdf   Lots more example photos and videos from real-world use here: http://philipbloom.net/2013/01/13/speedbooster/   A minor technicality- the device is actually a 1.09 crop (close enough :)).

I'm going to use the FS700 + MB SB to create sharper, full frame cat videos for youtube. That and test charts. Maybe combine the two concepts and shoot a Hello Kitty test chart.

According to the MB whitepaper*, the SB will increase image quality (measurable MTF) in the same way an extender reduces image quality. If accurate (in mass production- the math/physics makes sense), that means that not only is FF possible and -1 effective f-stop, but also higher IQ. I ordered the SB and and FS700 to test. If it works as advertised, the FS700+SB looks to be the most affordable/versatile pro camera right now (S35 and FF, very low light, multiple frame rates, super slomo, true 1080p pixels). C300 has better ergonomics and perhaps a more bulletproof internal codec, but has far less features and costs $16k.   * http://www.metabones.com/images/metabones/Speed%20Booster%20White%20Paper.pdf

Finally a worthy upgrade to the 5D3 (was considering the 1DC): MB SpeedBooster + FS700. My thoughts here: http://cinema5d.com/viewtopic.php?f=30&t=45529

12-core MacPro's are a great deal if one is looking for a fast, stable quiet Mac or PC (or both). I use my 12-core MacPro mostly as a Win7 workstation (+Quadro 5000 and GT120 (3 displays)). The Q5000 now works in OSX too, but only after booting (not in the ROM screen).   It's not possible to find a lower-priced 12-core for a PC (might be possible building from scratch, however the CPUs and dual-socket MBs are very expensive). If you factor in your time to get everything working properly, it's probably not worth it (I used to build all my own workstations from the best parts). HP and Dell make decent 12-core workstations, but they are not any cheaper.   If one is looking for the absolute fastest workstation, Boxx appears to make them: 16-cores, water-cooled and overclocked, with multiple GPU cards: http://www.boxxtech.com/Products/3dboxx-8920 If you have to ask how much, you probably can't afford them.   I can edit everything I need in real-time in PPro CS6 (except noise reduction with Neat Video)- so no need to render out to see what the final look will be (I only use real-time GPU effects) in most cases. After Effects is slow no matter what, so I don't use it (except for occasional experiments). I can also edit most everything in real-time on my MacBook Pro Retina in OSX Lion. Unfortunately (and amazingly), it's not possible to share PPro CS6 projects between OSX and Windows without issues (at least for medium complex projects), even when doing a clean archive/export. As others have noted, Win7 is a bit faster than OSX on the same hardware.

If we look at the trend of computer graphics (CG), the issues of sets, props, and makeup not working well at HFR and high-res won't be an issue when the cost of 100% CG becomes cheaper than shooting live actors and sets. It's been done before, but has been very expensive and the quality isn't quite there yet. When 100% CG happens, stereoscopic 3D becomes easy (basically free), everyone gets mo-cap'd, and all audio is VO (as with Gollum etc.). At some point even VO won't be needed- famous voices will be synthesized with an advanced text-to-speech system (with nuance controls). The final step is computer simulations will write and create the entire movie. This could be the peak of a golden age when analog, biological brains created content using physical cameras and live people.

A supersampled 4K sensor for HD/2K is excellent: that's what we need for true HD/2K resolution sampling without aliasing. 4096*2304 = 9.4Mpixels. Sony's 8.9Mpixels (effective sensor size) aren't enough samples for 4K without aliasing. To get proper 4K we need to sample 8192*4608 pixels: 37.7Mpixels.

Even if only one line of code change, some companies are very conservative and might take months of QA before releasing an update. Folks in the field will get early releases to help with testing. However, 6 months appears to be business timing related.

Based on ML dev comments, the actual resolution might be 1620x910 (also matches resolution tests we did (~810 lines of horizontal resolution)). Hopefully the new firmware will provide something higher than that. 422 would double the color resolution, which is helpful.

I suspect the code change looks something like this: Current code: #define DISABLE_CLEAN_HDMI_OUT New code: //#define DISABLE_CLEAN_HDMI_OUT It's already done and was used during initial testing and development. Hopefully the next change will be: //#define ENABLE_MARKETING_CRIPPLE_H264_COMPRESSION_QUALITY //#define ENABLE_RESOLUTION_REDUCING_GAUSSIAN_BLUR This would enable higher resolution and high quality H.264 output. The combination of blur and low quality H.264 output puts limits on post sharpening due to compression artifacts.

Release date April 1?

I was referring to this video: http://www.youtube.com/watch?v=fjtGenTSfX8 from this page: http://www.magiclantern.fm/whats-new/104-releases/143-second-alpha-for-5d-mark-iii

HDR looks really good and the complete lack of aliasing really helps this footage look like film.

Full frame or cropped? (not clear from that thread). Upping the bitrate of IPB + tweaking would be very useful, as would increasing the resolution before compression from ~1700 to 1920. C300 looks great at 50Mbps (MPEG2)...

What is interesting with this tech is that if you use a known camera preset when shooting ("Standard" for Canon DSLR's (they are working on other more useful presets such as Neutral, Faithful, etc.), then apply their preset filter, you'll get a decent approximation of the modeled film grain and color. This is a non-linear process, and it looks pretty good. Once processed into "film" normal CC and effects can be done for mood (as with analog/chemical film). I won't use the current iteration as it's not GPU accelerated in PPro (Windows): too slow. It appears to be GPU accelerated in FCPX (real-time).

@yellow- if the ML devs could report the resolution output before any up-rez to 1080p, that would be insightful when comparing to a chart. Also helpful would be a physics & optics explanation why a large sensor looks better vs. a smaller sensor. Exactly what is going on with the captured pixels, etc. People who don't know anything cameras respond very positively to large sensor images. In order to be 'filmic' and image must not have any aliasing- that's a key give away for a digital image. Phil Bloom's test shoot looks nice, but some shots have aliasing. The ideal camera has no aliasing. If that comes at a cost of perceived resolution- that is better IMO. In the video game and graphics card industry, we went to great lengths to produce anti-aliased images. Indeed as still images, the anti-aliased images look softer. However, in motion the anti-aliased images look much, much better. [url="http://en.wikipedia.org/wiki/Spatial_anti-aliasing"]http://en.wikipedia.org/wiki/Spatial_anti-aliasing[/url]

[quote name='cameraboy' timestamp='1347002913' post='17496'] @jcs sorry man ... but that image is way oversharpned ... u can get away with that when u shoot vistas but for faces not... and u can get 1000 lines from 5d3 if u put camera close enough .... testing resolution is precise process ... and all professional tests show no more than 600lines ... [/quote] What artifacts do you think you see from sharpening? The noise was added in post- perhaps you don't like the noise? (can be changed to suit taste). In any case, it is a very low light shot- not sure if it's possible with a camera providing 1600 ISO. Did you look at the video (no noise added, has reasonably fine detail)? A close up won't capture increased resolution (not possible to get 1000 TV lines by zooming/moving closer. If done with a chart, the result would be wrong (must align to markers)). The German site you referenced for resolution showed a contrast graph but no chart, likely with no post sharpening (which the 5D3 needs in order to show actual detail: many cameras can't be sharpened that much in post). They didn't post a resolution chart. Do you have links to professional chart shots for the 5D3?

5D3 skin, hair, clothing detail: [url="https://vimeo.com/39445020"]https://vimeo.com/39445020[/url] If you download the uploaded MP4, you'll see detail in the skin- pores, scars, wrinkles, stubble, single strands of hair is visible on head, detail in the sweater fabric, etc. Note also low noise in the video. Keep in mind this was recompressed for Vimeo upload, and thus detail was lost. [img]http://www.eoshd.com/comments/uploads/inline/16978/50495331a4318_Ellyn5D3LowLight.jpg[/img] Here's a still frame from 5D3 video shot handheld in low light (don't recall ISO, perhaps 5000+; see her pupils), F4 on the 24-105L (not the sharpest lens and not critically focused). The 5D3 did indeed reduce detail in this condition (part of internal NR), however simple noise grain was added in post to bring back texture (runs in real-time in PPro CS6). Not super sharp, but a pleasing, usable shot in low light. The images posted so far for the BMCC have a lot of fine noise- that's not real detail or resolution. Thus, charts help us understand actual detail and resolution capabilities of camera hardware. [EDIT]This forum won't show the full resolution image: full res image: [url="http:///www.brightland.com/t/Ellyn5D3LowLight.jpg"]http:///www.brightla...5D3LowLight.jpg[/url][/EDIT]. [EDIT2]Just watched Bloom's "Dungeness" demo (downloaded MP4 from Vimeo). It is indeed sharper than anything I can do with the 5D3 currently. The images look nice, and remind me of my Panasonic TM700 in bright light (also higher resolution than the 5D3). Thus, to my eye many shots have a video look, perhaps fixable with grading[/EDIT2].

We need to be clear when we discuss lines of resolution- typically that's TV Lines, or horizontal line pairs. 780 lines is less than I and Jason Greene measured. From our ISO 12233 tests I think it's fair to say extinction occurs around 850 lines (horizontal pairs), for 1700x956 resolution. The BMCC might do better in the ISO 12333 chart and the post-sharpened 5D3 might do worse in the Lemac chart. The only way to know is to test the cameras with the respective charts to allow a direct comparison. If someone mentioned 780 as vertical pixel resolution for the 5D3, where is the chart to support this? If it is really horizontal line pairs, that would be 1560x878 pixels- very close to 1600x900.

From this chart: [url="https://docs.google.com/folder/d/0B4KzkrvuLxMKNXBUemloakpWU28/edit?pli=1&docId=0B4KzkrvuLxMKS1VmTnlhRTRCWDg"]https://docs.google....S1VmTnlhRTRCWDg[/url] extinction of detail and excessive moire appears at G, indicating [b]1600x900[/b] resolution when all the photosites are combined via de-Bayering (2.5k is for the raw Bayer photosites- not the actual resolution of output image). This matches expectation of what we would expect from a 2.5k sensor. The 3.8k sensor of the C300 provides full 1920x1080 (and beyond depending on where one draws the line at extinction of details), again as expected from the math (Nyquist sampling theory). See this chart: [url="http://provideocoalition.com/index.php/awilt/story/quick_review_canon_c300_super35mm_lss_cine_camera/P3/"]http://provideocoalition.com/index.php/awilt/story/quick_review_canon_c300_super35mm_lss_cine_camera/P3/[/url]. It would be helpful to see an ISO 12233 chart to compare with other cameras (5D3, GH2, EX1, and others on the internet, e.g.: [url="https://vimeo.com/39536799"]https://vimeo.com/39536799[/url]) and/or a post-sharpened 5D3 shooting the Lemac chart.

If we apply a Gaussian blur to an image, it looks soft (as with a strong AA filter). If we apply a sharpen filter (convolution, increases micro contrast using local derivatives (changes in intensity)), we can get something close to the original detail back. It won't create new detail, but it will show existing detail more clearly. Thus, we're not increasing resolution from ~600 to ~890- only enhancing information that is already there. Jason already did something similar with the 5D3, T3i, and EX1: https://vimeo.com/39536799

For non-German speaking folks it is inaccessible. 5D3 footage scaled down from 1080p to 720p loses detail and resolution: 600 lines pairs is not an accurate measure of usable horizontal resolution (that appears to be your take on their graph). Let's compare ISO 12233 charts and make our own decisions. The 5D3 is indeed a bit soft before sharpening in post (with in-camera sharpening turned off): resolution should be measured after post sharpening to counteract the strong AA filter (as Jason did here): https://vimeo.com/39517721. Earlier I had performed a chart test along with a scaled down still: Jason has a printer which allows for a proper test: our results matched pretty well. When doing a comparative test of video vs. a still (for DSLRs), one can get a good idea how well the camera is performing with their own equipment (and not blindly trusting what is said on the internet). A full size ISO 12233 chart can be used with any camera- anyone can test with their own gear.

Sorry, Andrew, I need to see a line chart. You are quoting 600 lines for the 5D3 and 1000+ for the BMC with no objective evidence (posting a link without an ISO 12233 line chart isn't helpful). Mathematically, we need 3.8K samples for 1920 pixels without aliasing. The BMC only provides 2.5K samples, or 1250 pixels without aliasing. However, in practice extinction of details happens later. Thus, a line chart for both cameras will allow a direct comparison of actual performance. That time adds up in terms of production time and cost. Extra complexity also increases the chance for a mistake (sometimes due to software / OS bugs). My thoughts are the BMC camera is a dream tool for those wanting to learn extended post skills. I would prefer to get something higher end which simultaneously improves quality and maintains or reduces production time (true pro audio built in, NDs, image stabilizer, etc.). Again, my personal preferences.