jcs

It's a common misconception regarding the Speedbooster and F-stop: the Speedbooster increases the T-Stop not the F-Stop. F-stop = focal-length/aperture-diameter: the Speedbooster doesn't change this. It simply scales down the image circle. The T-Stop increase is relative to the same lens without the Speedbooster on the same camera (e.g. the regular Metabones adapter). Notice how the bokeh doesn't really change when comparing the same lens on full frame to Speedbooster on S35 (size change is due to 1.1 crop (Speedbooster isn't quite full frame)):

Right- a m43 12-35/2.8 on m43 will have the same bokeh and light gathering performance as a FF 24-70/5.6 on a FF camera (not very fast, not much bokeh compared to F2.8 on FF). He demonstrated this effectively in the examples in his video.

The circle/particles/area comment was to show that a larger diameter bucket will capture more water per unit time vs. a smaller diameter bucket. Pixel density doesn't matter with respect to the math and real-world results when using the math to predict what will happen with various lenses, sensors, ISO, and f-stop settings. ISO or gain is calibrated to the sensor- this is why the details of the sensor design don't matter. If ISO is increased from 100 to 200, we've doubled the gain and image brightness in a known way, regardless of anything else because ISO is calibrated to that sensor. Is there variance in camera manufacturer ISO calibration? Sure, however his examples match the math. Using any of items 1-3 above as shown in the video, is there a real-world example that shows the math is wrong?

My understanding of ISO is it's calibrated gain, which goes up in powers of 2, where each value doubles the gain: 1, 2, 4, 8, 16, 32, 64, etc., shown as 100, 200, 400, 800, 1600, 3200, 6400. Sensor details don't matter if ISO is correctly calibrated. Some cameras show gain in dB. In any case, I haven't seen any counterarguments or errors in the math for the above 3 items when used as a tool to understand fov, brightness, and bokeh (and noise to some extent). Folks have issues for other things he's said in the video, however I'm focused on the math and the examples which support the math: it works for the examples shown so far. Regarding a smaller diameter bucket capturing less water vs. a larger bucket: a greater diameter bucket has more capture area and thus will capture more water per unit time for uniformly distributed rain particles. The depth of the bucket only determines how much water the bucket can hold till it overflows. In Photoshop create a new file and add noise. Draw a circle in it. Now draw a smaller circle inside the first circle. This represents a large and small bucket capturing rain particles as a snapshot in time. Which circle incloses more particles?

The beauty of math is it stands on its own- where is the math error? The examples match the math prediction, so it's a useful construct. Samuel's 2011 DOF/FOV calculator does the same thing and matches the video. Separate from anything else said or shown, are there math errors for items 1-3 above? Is there an example of the math not working for any case?

With respect to the simple math and examples, where are the errors? If it's not possible to point out an error with this math or images, what he says in the video is essentially correct: Multiply focal length by crop factor: well known and the point of the crop factor for full frame reference. Multiply aperture by crop factor: this matches his examples for apparent brightness and bokeh. Multiply ISO by (crop factor)^2: this matches his example for apparent brightness and makes sense based on what ISO really is: gain; the square term makes sense for sensor area, and works for the examples shown. Samuel (flaat profile creator) did tests and includes an FOV/DOF calculator which exactly matches what is in the video. He made this post in 2011: http://www.similaar.com/foto/doftest/doftest.html

Andrew- let's start with his math: where are the errors? He shows example images which match his math: what's wrong with the images?

Haven't watched the whole video; he makes a few mistakes, however his math makes sense so far and the images and settings match the math predictions. So far, he is generally correct. Math followed up with experiments and examples that match the math is a strong argument. Any challenges should be made to his math and the experimental results.

Some interesting locations for sure. I've been reading this book: http://www.amazon.com/Shoot-Video-That-Doesnt-Suck/dp/0761163239/ and one of the key points is to always tell a story, no matter what (even a test video), otherwise folks won't watch your work. Thus, write a beginning, middle and end, decide on the 'hero' of each scene, then shoot it. Just a little bit of planning goes a long way. For example, you could start off talking with friends about scouting locations (intro characters and set the stage for what is to follow: the beginning of the journey). Then film the locations and talk about what is interesting and useful in each location. Then wrap up back at home base with a quick recap of what areas will work for what purpose. To get more creative, you can have something happen while scouting locations: something scary usually works: an unknown race of mutant carnival workers crossed with escaped mental patients have taken one in your group. Now you have to rescue him/her. Only one or two make it out alive to tell the story. The hero(s) learn not to mess with mutant mental carnies unless packing sufficient heat and having backup available.

If the GH4 files can be loaded by Resolve (Lite, free), you could use Resolve to transcode 4K to 1080p if you don't want to upgrade to CC.

Thanks for the tips Michael- I'll give them a try!

It was shot at 50fps at 1/2 vertical resolution (ML option). Then stretched and played back at 25fps in post. Twixtor is a plugin for interpolating frames for slow motion.

Right now, the GH4 shooting 4K (even if delivering 1080p- downsampled in post it's the sharpest/most-detailed currently available in that price range).

While you can sharpen the 5D3 footage a bit more and turn down C100 sharpening (or off, even Gaussian blur it slightly in post), if you sharpen too much it will start to look digital (artifacts). For some of the up close portrait-like shots, the softness of the 5D3 is pleasing in your wedding video. Did anyone notice a difference in cameras for your video? The cameras cut together pretty well. If you used an FS700, you could use a SpeedBooster to more closely match the 5D3 FOV & DOF. The C100 has an advantage of better color processing vs. the FS700, especially for skintones (I've tweaked the color profiles for the FS700 and am more careful with lighting to help better match the 5D3 color). If you want to get the sharpness and detail a lot closer to the C100, and far surpass it in color processing possibilities, shoot 5D3 14-bit RAW. Here's a wedding shot in 100% 5D3 RAW: Details: http://www.dvxuser.com/V6/showthread.php?322289-Wedding-(RAW) Whether the extra time and effort shooting RAW is worth it for weddings is another question.

Starting at 3:45, see ALL-I compared to IPB, especially the "10" section on the lines (horizontal or vertical): There's more detail in IPB (from this thread ~2 years ago: http://philipbloom.net/forum/threads/intraframe-all-i-and-interframe-ipb.987/ ). Here's an example I shot handheld (not a locked off shot- there's motion): http://www.dvxuser.com/V6/showthread.php?279229-Canon-5D-Mark-III-IPB-contains-more-detail-and-has-less-artifacts-than-ALL-I/ , as well as another discussion: http://www.dvxuser.com/V6/showthread.php?278746-ALL-I-vs-IPB-IPB-Wins When policar posted a still frame grab of fast motion in 1080p in ALL-I vs IPB, the ALL-I still frame did look less smeared. However, during playback this kind of motion detail may not be visible. It's really important to test for yourself vs. simply following the advice of others. I did find that 720p60 always look better with ALL-I on the 5D3 which I added later to the thread. ALL-I vs IPB is a popular topic. The common argument is "ALL-I must be better because the bitrate is higher and the compression is less". However, in real-world practice, most people can't see a difference. There are many tests out there showing this: Again, for motions shots, if there aren't any objectionable artifacts for IPB (including excess high-frequency filtering / detail blur), then IPB can be a better choice to save disk space. While I haven't done a test for medium or fast motion, for low motion, the 24Mbps FS700 AVCHD provides more detail than 5D3 RAW (in the same way the GH4 provides more detail than 5D3 RAW: '?do=embed' frameborder='0' data-embedContent>> . I purchased a Nanoflash external recorder for the FS700, however I've never used it beyond testing as the 24Mbps internal codec's quality outweighs the extra weight and complexity of an external recorder's marginal image quality increase (again, except in motion shots and certain scenes such as water, etc.). If you have an example or know of one showing ALL-I clearly doing a better job than IPB, it would be helpful to see under what conditions this can happen, e.g. what is the motion threshold where ALL-I starts looking better than IPB (and is it visible during playback vs. pixel peeping a still frame).

Huuow- haven't used an LED Fresnel yet. Sounds like tungsten is the most cost effective at the moment. While not a Fresnel, the Alzo-3000 looks decent: https://www.youtube.com/watch?v=bHeImPqjoCM, but does have some fan noise (supposed to be fixed next production run): http://www.learningdslrvideo.com/alzo-3000-led-light-review/ (and colors aren't as accurate as his fluorescent lights). This has a good review: http://www.amazon.com/Fotodiox-Daylight-Fresnel-High-Intensity-Television/dp/B00CN3032W/ref=sr_1_8?ie=UTF8&qid=1399876188&sr=8-8&keywords=fresnel+led+light

Had to do a quick shoot and left 4 of the Cree 100W bulbs in the key light. Shot at 23.976, 1/48: no flicker at all was visible. CFL and 4 kinds of LEDs were used (all around 5000K): worked well (no grading was needed in post). The bicolor LED panels use two kinds of LEDs, 1/2 warm (incandescent), 1/2 cool (daylight). When using 100% of one light color or the other, only 1/2 the LEDs in the panel will be active. So, a 600 bulb bicolor light is the same brightness as a 300 single color light when using pure warm or pure daylight LEDs (some panels don't allow you to control the color banks individually (they automatically blend color levels as the dial goes from warm to cool)). I use a smaller Z96 light on camera. It's never on the camera unless I'm shooting in the dark. The YN-300's a bit larger- it can be used on camera but you might find something in the Z96 or YN-160 size more suitable.

The 100W (18W used), 1600 lumen Cree 5000K LED lights were $21 each. They are super bright, and look pretty good. When recording at 24fps, they look OK. However at 240fps (Sony FS700), they flicker. I tested the Yongnuo YN-300, YN-600, and F&V R-300- no flicker at 240fps. The old 120W (45W used) 5100K CFLs don't flicker (though some make noise & buzz). Here's a 60W Cree tested (no flicker at 24/30fps): Those looking for budget lighting, I'd start with the Yongnuo YN-300: 2280 lumens for $67 http://www.amazon.com/Yongnuo-Professional-300pcs-sheets-Camera/dp/B00AZFE5DS/ref=sr_1_1 and this power supply (in addition to Sony NP-970 batteries for non-AC use), $25: http://www.ebay.com/itm/110V-220V-AC-Adapter-Power-For-SONY-NP-F970-F960-F750-F550-YN-300-II-YN600-YN160-/291005963183 Based on specs only (not tested), 2 Cree 5000K lights are 3200 lumens for $42 + $20 for 2 clamp-on bulb holders = $62. The YN-300 with AC adapter is $92, however it's also dimmable, camera mountable, doesn't flicker at high frame rates, and can also run on batteries. The YN-300 II and YN-600 are "bicolor" and more versatile, however they are effectively 1/2 brightness when using the daylight LEDs (I'm using lights in the daylight 5000+ K range). I'm going to keep looking for LEDs that don't flicker at high frame rates to replace the CFLs. This info is helpful: http://www.davidsatz.com/aboutflicker_en.html

Going to Home Depot to pick up some 100W 5000K LED bulbs to test. Not clear if this is the datasheet for the LEDs used; the spectrum appears to be reasonably continuous: http://www.cree.com/~/media/Files/Cree/LED%20Components%20and%20Modules/XLamp/Data%20and%20Binning/XLampXTE.pdf

I was curious why daylight and tungsten was easier to deal with, especially for skin tones. Looking at the spectral distribution of light, tungsten and real daylight provide a continuous spectrum of light, while fluorescent/CFL produce highly spiked light- not continuous. The good news is that the latest LED lights come very close to tungsten/daylight for continuous spectral light. I've got some low-cost CFLs in 2x3 and 2x2 light rigs: I'm going to replace the CFLs with LED bulbs at around 5000K (probably Cree- thanks for the tip FilmMan!) to match the other ~5000K LED lights I use (currently Z96, YN300, YN600, and F&V R-300). An LED spot in a clamp-on work light will also be handy. http://www.cs.cmu.edu/~zhuxj/astro/html/spectrometer.html http://www.popularmechanics.com/technology/gadgets/tests/incandescent-vs-compact-fluorescent-vs-led-ultimate-light-bulb-test#slide-1 (see the spectral charts) (CFL can work OK, but I had to use a tuned camera color profile and make sure WB was set carefully, otherwise trying to fix in post was time consuming).

This plugin can probably help remove ACR flicker for 5D3 RAW footage: http://www.digitalanarchy.com/Flicker/main.html

Thanks for the tip elgabogomez!

If no motion artifacts, IPB is more efficient and tends to work better. The same is true for the 5D3.

It would be nice if Primatte and Keylight were accessible in PPro. Ultra is real-time in PPro and does a very good job if the green screen is well lit. Here I lit the green screen better which helped with keying: [we're moving in the direction of humor ;)] The green screen tricks I was referring to were at the code level (custom C++ and GPU shaders): we removed backgrounds without a green screen. Can't discuss the algorithms though it would be nice if they were implemented in commercial keyers. Regarding the quesadilla- I chose those ingredients based on taste and nutrition. Coconut oil adds healthy fat to help reduce the glycemic index of the tortillas. Cayenne helps reduce potential inflammatory properties of cheese. Corn was chosen as it doesn't contain any wheat and is generally less processed (ideally non-GMO). If quinoa tortillas exist, that would be another good option (or rice since some folks have corn allergies). Reviewing grapseed oil- it's high in omega-6 fatty acids (inflammatory) and may contain hexane from processing (100% natural oil on the label unfortunately doesn't really mean 100% as allowed by the FDA). So the oil I recommend the most is organic cold pressed coconut oil such as Barleans: http://www.amazon.com/Barleans-Organic-Virgin-Coconut-Ounce/dp/B00ATW2QII/. Coconut oil has the least omega 6: http://authoritynutrition.com/optimize-omega-6-omega-3-ratio/ Olive oil isn't a bad choice, however I prefer the taste of coconut oil. Thanks for the comment regarding oils- I had purchased a new brand of '100% natural' grapeseed oil and had not researched it.

.