4K camera purchase for Kinograph v2 testing

Hi all. I’m ready to do some basic testing with a 4K camera and I need your help.

Firstly, let me explain why I want to buy a nice one instead of something cheap like the Raspberry Pi camera (which I already have and will most certainly be testing with eventually). I believe it is best to start with something that is going to give us the best possible outcome, and then downgrade as necessary in subsequent tests. This will ensure a benchmark of quality that can be relied upon, and to which we can compare other results.

Here’s where I need your help. I believe I have found the camera I would like to purchase and would like your opinion.

Flir Blackfly BFS-U3-123S6C-C

Here was my decision “funnel:”

  1. A resolution of 4096 x (whatever). 3840 × 2160, is the dominant 4K resolution in the consumer media and display industries. 4096 × 2160 is the resolution used mainly in digital cinema production. Since our end user device is a monitor/display, I believe the former is our target resolution for testing. If you want more, you can spend more. By scanning at 4096x(whatever) then the user can safely crop out perf holes, soundtrack, etc and still maintain the 3840x2160 resolution (see below for a discussion on the 2160 number though).
  2. Of the cameras within the Blackfly family with that resolution, I chose the top three that had >24fps, a high dynamic range, and at least 10-bit depth. Here is the side-by-side comparison of those three models.
  3. Of those three, only one has greater than 2160 pixels in the vertical axis. Because it can be advantageous to overscan (to account for jitter, etc), I believe having those extra pixels is important. Then the user can crop in the vertical direction and still maintain a true 3840x2160 resolution for 4K monitors/screens.

The one remaining camera from that funnel of decisions is the one listed above.

What say you? Are my assumptions solid (resolution, overscaning, etc)? Is there another factor I’m missing?

CC @johnarthurkelly @Doc @Mike @Udayarangi @fabutch

PS one thing to consider is that some have complained about needing to convert the images produced by the Flir cameras into DNG as a separate step. See this thread.

As for the resolution you should base that on what you’re trying to achieve. 4096 pixels horizontally won’t help much for scanning film if your vertical resolution is just 2160 pixels. I’m personally using the 7.1 MP-version which has a resolution of 3208 x 2200 pixels since that was closest to the ratio that I wanted to capture.

When scanning 35 mm-film you even have “dead space” with that resolution on the left and right side if there’s minimum overscan:

So if you really want to do UHD-resolution afterwards on a native source you’ll have to go for the 4096 x 3000-version (12.3 MP) as the other two won’t help you there. You also have to check whether or not it’s able to capture 24 FPS if you’re using that maximum resolution. According to the data-sheet the 12.3 MP-model will only do 19 FPS when capturing in color at 12 bit: Performance Specification Document

1 Like

@Doc makes a good point about the amount of the sensor used for a given frame ratio - of course, we’re always going to be fighting against the aspect ratio of the image on the film in the scanner unless we’re matching it + overscan for sprocket holes and soundtracks, etc. Which is where the idea of oversampling is so very attractive, but doubling our sample vertically and horizontally quickly puts the camera choices in the range of VERY very expensive, and even then, FLIR only has one camera that is -almost- 8K (in pixel count, not aspect ratio, 6464 x 4852).

It’s just like the old triangle of good-fast-cheap only in this case it’s temporal resolution-spatial resolution-bit depth, most of the options available seem to sacrifice at least one of them to max out another. If the machine can sacrifice 24fps (and the imaging mode @Doc found in the spec sheet is the one being used) then the model you linked directly @matthewepler is pretty great!

2 Likes

The final display aspect ratio shouldn’t be a factor in this decision at all, in my opinion. The purpose of the scan should be to capture the full frame, at its native aspect ratio, for processing later. For that reason a 4:3 sensor makes the most sense as it fits closely to almost all film gauges. Techniscope would be an exception, but even anamorphic film would be squeezed into a 4:3 image area and then digitally unsqueezed. The display aspect ratio is something to think about after you’re done grading/restoring the film.

The typical size for a 4k scan is 4096x3112, unless it’s a wider format like 2-perf, or Super 16.

These are great responses. Thank you, everyone. I’m hearing in all three of them that the 4096x3112 is a good ratio. Thank you @Doc for finding that stat on the fps vs bit depth. Since we aren’t supporting real-time sound acquisition at this time, I think that’s an acceptable exception for now.

Similar to Doc’s link, here’s the frame rate table for the model I posted specifically above. You’ll see that at 12-bit, it can go up to 16 fps at full resolution.

Thank you also for the discussion regarding overscanning/frame ratio/resolution. Illuminating! Pun intended.

I just checked the contracts with my two confirmed pilot partners and neither require 4K. So I’ll be putting this off so that I have maximum budget left for building the machines for the partners and additional experiments.

But the conversation is still highly relevant and valuable. Thank you all!

I’d stick with the imx253 imager for 4K as those are the ones that are used in the commercial scanning machines (Lasergraphics, Filmfabriek, Kinetta, etc) or Pregius S which would be the imx535 (but that’s a moot point since it’s not available in a Blackfly S camera). We also already know that the BFS-U3-123S6C-C is a good camera, if you were considering the BFS-U3-122S6C-C as a cheaper (and slower) alternative it’d be best to get Flir to send test cameras first so you can see if there’s a difference in quality.

They’re not quite “full 4K” as Perry has mentioned with the vertical resolution being a bit below 3112. But there’s something he didn’t tell you about the typical 4096x3112 scanners :wink: which is that most of them are line-scanners meaning they have cameras with a resolution of 4096x1 or 4096x3 etc, they take many photos and stack them vertically to create a frame, and on many of them you can’t scan the perfs you only get the film area. The Sony Pregius sensors are great quality, very fast, and cheap as well so really anything still using linear array CCD imagers is old-hat now. 4:3 sensors do make the most sense, but there is compromise to be made since some cameras simply come in the sizes offered like the imx530/540 at 5320x4600.

The imx530 or imx540 would be the choice if you want full 4K, although it will be slower fps capture, larger file sizes, and shallower depth of field. For those reasons some people feel the imx253 camera offers better value overall even though it costs a bit more than the newer 5K cameras.

Resolution isn’t everything, unless you have really good negs you’re not getting 4K detail out of most of them anyway. You can scan most film at 2K and bump the resolution up to 4K and no one will be able to tell the difference between a native 4K scan. Until a few years ago 2K was industry standard for scanning. The resolution has increased because the tech is a lot cheaper now, but the film material has a finite resolution and for most films that resolution isn’t as high as we’d like to believe. Unless you need to scan in 4K or 6.5K the slightly lower res cameras will offer better bang for the buck and produce files that are easier to work with. What I feel is more important than resolution is to get a good quality capture - minimal sensor noise, no clipping, good dynamic range. If we were talking 8 years ago we would be compromising a LOT over those options, but now I feel we’re at a point where we can confidently say we don’t need need to make those compromises any more.

1 Like

Well, kind of. No currently available (new) scanners use line arrays any more except for the Scanity, but who cares about that heap of overpriced hardware?

4096x3112 is based on the 1.31 (4:3) television aspect ratio. 4096 is the generally accepted width of a 4k frame, and the height is whatever you need to make up the aspect ratio. I’m not aware of any scanner that uses a 4096x3112 sensor but generally files are output to this size even if the aspect ratio of the frame is a bit different. 4096x3112 is an aspect ratio of 1.31, or 4x3. But film is all over the place in terms of aspect ratio. To keep things simple, since 99.99% of the time we’re scanning with some overscan, we always make a 4096x3112 file for the client, and do any cropping later. That means a regular 8 and a super 8 scan would be the same pixel count. even though the super 8 frame is ever so slightly wider than 8mm.

Hi Matthew,

Ref. Flir BFS-U3-123S6C.

  • I assume that you will have a sensor increase i.r.t. what you have now in your current set-up. The 123 has a sensor of 1.1”. This means you have to adapt the focal length (distance sensor to lens). Also your working distance might be impacted. How much I do not know. I would determine it based on experimenting.
  • The 123 has 3.45 micron pixel size. I think it is good not to have smaller pixels because noise is enhanced. I bought the BFS-U3-120S4C (4K, 1/1.7 sensor) initially (was also advised by Edmund Optics) which has 1.85 pixel size base on the assumption that smaller pixels would give better (sharper) images. But to my surprise that was not the case. That’s why I moved to BFS-U3-50S5C (2K, 2/3” sensor).
  • It has a 71 dB dynamic range which is almost the same as my 50C. As far as I can judge this is a high range. But my 50C still lacks dynamic range. I.e. it easily get over exposed of detail becomes hidden in the dark. I am still working on that .

Gr,

Mike

1 Like

I just meant most existing scanners already in use. That’s how the typical 2K and then 4K scanning standards were defined, the area scanners are the newer machines (although some of them are quite old now!) Many of the old HD telecine machines could be upgraded by putting a 2K or 4K line sensor in and scanning to DPX for example, I’ll bet that a lot of companies 10-15 years ago found that option a lot more affordable than buying a new digital scanner. The current Scanity sensors are 96 lines high.

The CCD-based line scanners have their fans, there are some users who want to stick with trilinear CCD and avoid demosaicing. Of course they have their own issues.

The overscan of some of the machines is great, and it gets you beyond 4K as you say. Perfect for cropping the film to its desired area and stuff like that. So for people that are shooting film, the 5.3K Sony Pregius S is perfect really, or the older 6.5K imager that’s already in wide use.

Hi !!
Very good camera…but the price !!!
I’m using the Lumix gh4, slow without global shutter but the image is very good !!!

Troels

1 Like

4K isn’t really necessary for 16mm, is it? I think the highest resolution you can get from 16mm is 2K, anything above that and you’re not really adding any, if any, extra detail, correct? I’ve been looking through the list of Blackfly cameras trying to see which would be satisfactory for my use. The vast majority of my collection is 16mm. I do have some 35mm trailers, though.

So when it comes to the camera @matthewepler mentioned in the first post, would that be good for someone like me, or does anyone else have any suggestions? I’ve had this one bookmarked, but am not sure if it would be good enough.

@DPXTIF what is your use case? Public screenings? Scanning for an institution? Yourself? In other words…how would you order some of the following in terms of priority for your scanning?

  • speed of the Kinograph to scan your film. This will determine what fps you will require from your camera and what kind of processing power you will need for your PC
  • size of the final product when screened (giant theatre screen vs. embedded video on website, vs 4K TV., etc)
  • budget!
  • dynamic range and bit depth…how concerned are you about getting every bit of visual information out of each frame? Or do you just need to see what you have so you can prioritize professional scanning later on?

There are probably more I’m forgetting at the moment, but you get the gist. If you don’t know the answer to all of these, that’s okay too. Just envision what you want someone to say after they see your scanned film. Who is that person? What do they care about most? As a former designer, I usually start with the user and then add on my personal preferences if there’s still budget :slight_smile:

@matthewepler I plan to use a Kinograph simply for my own purposes. As I’ve mentioned before, I’ve been collecting film for nine years this October, and ever since then, I’ve been waiting for something that could scan my prints in high quality. This project can provide that. Here’s my responses to the priorities you listed:

  • Speed isn’t necessarily a factor for me, but I prefer something that can scan above 15fps. I do have a computer which can handle 24fps, though. It’ll be used to scan the prints I have with ease and will allow me to edit the scanned files easily in post.
  • I usually view my transfers on a computer screen or phone. There’ve been times when I’ve looked at some transfers on a 1080p TV just to see how the quality looked.
  • Budget isn’t necessarily a problem for me. I think the most I could spend on a camera would be $2K, though that may change over time.
  • I am very strict about quality. The previous transfer methods I utilized were only temporary and provided video files that were decent in quality, but nowhere near the quality you’d get from an expensive film scanner. I plan to do quite a bit of work in post, so something that preserves as much visual information as possible would be ideal.

I’ve used two methods of transferring prints. The first method utilized a modified projector with a five-blade shutter, a CCD camera, and an LED light source. I captured the film in real time. The resolution of the camera was 480p and compression made the capture look slightly worse. Small text was hard to read and exposure wasn’t that great, especially if I was transferring a faded Eastman print. The color levels would be oversaturated and white balance was not great. The second method utilized an iPhone 7 Plus by pointing the camera into the projector’s lens and adjusting the AE/AF lock until flickering went away. This also captured film in real time. The quality from that was actually pretty good. Again, nowhere near as good as an expensive film scanner, but for me, it was satisfactory enough.

The collection largely consists of 16mm (160) with a few rolls of 35mm (42). I’m not really concerned about the 35mm prints I have; they’re all movie trailers. Transferring 16mm is my main focus. I’ve got a few prints of old TV shows that haven’t been seen in syndication in probably 25+ years and/or are not available on DVD. I’d really like to get those scanned and cleaned up somewhat. Everything else you can easily see on channels like MeTV, or on DVD. However, these are old syndication prints, some with commercials. I usually focus on collecting syndication prints, with a few network prints here and there. Why collect shows that are already on DVD? Well, why not? It’s a great experience to watch, say, an episode of The Andy Griffith Show being projected from actual film, rather than watching a digital file from a DVD.

There are a few people who would like to see some of my prints once I get them scanned, mostly fellow collectors or people who are curious about what prints I have. I hope that when they see my prints, they’ll have the same reaction as I did when I saw one of my prints transferred with a Lasergraphics ScanStation: “Wow. That looks amazing.”

1 Like

That’s very helpful. Thanks for explaining. From what you said, I think the 2K camera we’re using would be great for your needs. Of course, it’s all subjective. But it doesn’t sound like you need 4K to me so anything with at least 10-bits (preferably 12 if you can get it) of color depth and a high dynamic range then you’re golden! You should be able to get something below your $2K budget, especially if you don’t absolutely need 24fps. Good luck!

If you end up with something you don’t like, chances are someone here will buy it off you (as long as its decent).

The simple answer is no, but because of the Bayer problem you should scan higher resolution than 2K if you can. A 2K camera has 2048 sensor sites (pixels) across the imager and around 1556 vertically. However the Bayer filter sits on top, this reduces the actual number of red pixels to 1024x778, the same for blue, and green has twice as many pixels (they’re offset from each other so you could describe it as either 2048x778 or 1024x1556). 2.5K or 3K are acceptable choices. The 3K one you have bookmarked is an excellent choice.

What I’d suggest though is that we get you a cheap camera you can start with as an expensive camera isn’t going to get you a fantastic scan without the rest of the build being up to spec. Just working on things like getting the light even can take some effort and you don’t need your production-quality camera for that.

2 Likes

This is an old notion based on outdated information that simply will not die on the internet. Way back in the day (25+ years ago), when 4k sensors were outrageously expensive, and the notion of having a multi-format scanner was completely outlandish, manufacturers decided “4k is enough” for 35mm. because scanners that could do 35mm couldn’t refocus the optics to fill the sensor frame for smaller gauges, something like 16mm was a crop in the middle of that sensor at - wait for it - 2k. Thus it became gospel that 35mm = 4k and 16mm = 2k, and Super 8 = HD, which is all complete bunk. Also worth mentioning, those numbers were floated before the current generation of film stocks were released, which have finer grain than the stocks available. Mid-90s EXR and Vision1 and stock was kinda meh.

We have a longish article on our web site addressing why high resolution scans make sense for even 8mm film, but in a nutshell:

  1. The grain IS the image, and you want to resolve that grain as best you can to get the sharpest image. Grain is not the enemy!

  2. If you scan at 2k and display on a 4k screen, you have to digitally scale that image up 4x (twice in each direction) to fit the 4k screen. This will always soften the image.

  3. The sharpness of what’s on the film determined by the quality of the optics, how well exposed and focused the camera was, how steady the camera operator was, and the fineness of the film grain. The scanner should assume best case scenario, not lowest common denominator, and scan at the highest resolution possible, in order to faithfully reproduce the film that’s running through it. If that film is a little blurry, it should reproduce that blur. If it’s tack sharp, it should be capable of getting that too.

3 Likes

I was of the opinion that 16mm only had so much detail that if you scanned it higher than 2K, you wouldn’t be really adding any extra detail. It seems that everything I’ve read has made the same points that you just refuted, so I came to the conclusion that all of that was true since I didn’t see anything otherwise.

I’m gonna have to give that a read! I’m always looking for new information about scanning and resolution.

1 Like

Well, let’s say at least “it depends”. Certainly film grain has specific aesthetics for a lot of viewers/content creators - so much, that nowadays artifical film grain is added to perfectly grain-free electronic footage to make it more “film-like”.

On the other hand, film grain was and still is a technological nuisance. Film grain simply interferes with the normal perception of a movie. There is a reason why the old grainy film emulsions were phased out over the years in favor of less grainy film stock.

The issue is most noticable in small formats like Super-8 formats. Look at the following example I posted previously elsewhere. To the left is the original scan, with clearly noticable film grain all over the place. To the right, film grain was mostly removed from the original film stock, enabling the recovering of small image details otherwise lost in the spatio-temporal noise we commonly call “film grain”:

If you nowadays notice film grain in the cinema productions, it’s mostly an artistic decision.

The actual resolution of a camera/film combination is a function of the MTF (modulation transfer function) of the camera’s lens and the resolution capability of the film stock used. The later one depends strongly on the exposure - bright image areas have a finer resolution, while darker image areas do feature quite noticable film grain. The reason for that is pure physics with a little chemistry: the most light-sensitive crystals in the film emulsion need to be large in size in order to react even to tiny amounts of exposure. During the development process, these crystals are washed out and replaced by large, randomly arranged patches of dyes - which lead to perception we know as “film grain”.

Putting all limitations of the camera optics and the film stock used together, I would certainly conclude that the image information of a Super-8 film is comparable to the HD format - if all factors are working in your favor. Use Agfachrome stock instead of Kodachrome and a generic-brand Super-8 camera instead of a Leicina or a similar high end camera, and the achievable resolution will drop noticably.

Caveat: if you want to recover the image information obscured by the film grain (as in the above example), you most certainly want to have a little headroom while scanning/postprocessing - so I recommend at least 2k scan resolution for Super-8 material as a base resolution.

To make my point clear - I am talking about the raw image information the camera was seeing at the time of the recording and how to preserve and present this with current technology. I am not talking about the aesthetics of seeing every tiny bit of flickering color patches introduced by the film stock on a 4k screen in the hope to come close to the original viewing situation, i.e. viewing the small format film through a projection system in a darkened room. We are not there yet: no display available to mere humans comes even close to the dynamic range a Super-8 projector is able to achieve. Once we have these (and some other stuff like a full HDR workflow and a corresponding HDR delivery scheme) available in a few years, we might come back to the issue of faithfully digitizing and reproducing all the grainyness of the original film stock, letting just - as in the old days - our human visual system do the spatio-temporal integration to see through the grain (noise). I’ll bet it will require at least 8k scans for a Super-8 movie to record all the details of the film grain various film stocks are able to generate…

1 Like

I sometimes scan 16mm high con film (ASA 6-25) for titles and other oddities, you would be very impressed how much resolution is on each frame. The grain is so fine it’s hard to focus the scanner camera on it. 50D is also very sharp and a 2K scan would be disappointing. The more you go up in ASA (100,250,500) is where grain gets thicker. Of course if you are scanning an old dirty 16mm print there it won’t be so sharp but a 4K scan will avoid any upscaling as @friolator explained.

Here’s a frame from a 16mm chart I shot and scanned (in 4K) on 7234 last week. Not sure if the forum accepts 4K jpegs or downscales… here it is.

EDIT: forum did downscale to sub 2K so if you look at my image in fullscreen, that’s exactly what would happen if you watched a 2K scan upscaled on a 4K screen…blurry, in 4K it’s super sharp.

2 Likes