[Help Request] calculating PC requirements

because the grain is the picture. Resolving the grain is kind of the point of a proper film scan.

FWIW, we scan a lot of home movies for our customers even though our primary market is archives. By far the most popular package for home movies is a 4k flat scan, along with a lower resolution (HD) MP4 for immediate viewing. Check out Errol Morris’s film WORMWOOD on Netflix. We scanned all the 8mm home movie footage for that film at 5k, full ovescan, and it looks fantastic.

There is more image than people think on small gauges, and storage is cheap. You can get a 10TB drive for just a couple hundred bucks. Also, with 4k screens being pretty much ubiquitous now, if you scan to a lower res you’ll have to scale up to fit that screen, which is going to soften the image.

Just curious, do you plan to archive/deliver your R/S8mm in 4K? I also realize that some S8 is in really good shape and shot on really nice fine-grain stock, so if any flavor of 8mm warranted 4K, that probably would.

Super 8 is actually worse than R8 in most cases. the cameras were cheaply made, and the film tends to move around a lot in the gate, causing motion blur. R8 cameras were mostly based on solid, all metal 16mm mechanisms, and tended to have better lenses.

… - well, I guess that depends on what you want to achieve. In analog times, film grain was mostly an annoyance people tried to avoid. There is a reason a 70mm film has a different visual impact than the S8-copy. Over time, people got adjusted to the grainy film look and film makers started to utilize the film grain as an additional dimension of story telling. Your reference “WORMWOOD” on Netflix is a prime example - by using the old film stock, even showing the sprocket holes, the film maker marks these sequences as “old” to the viewer. So in this case, grain is clearly used as a means of story-telling. But that is not always the case.

In the digital age, grain as well as camera noise tends to massively increase the bandwidth requirements for storage, transmission and processing. Depending on your goal, that might be of no concern. Certainly, if you are going for archival storage (and that is probably the case for the Kinograph-project), you might want to scan at the highest resolution possible. However, most of your budget, specifically your time budget, scales approximately with the square of your scan resolution. GPU-processing won’t really save your day here.

Actually, basically everything scales with your scan resolution: the price of your camera, which lens you need to use, the rigidity of your machine, how good you have to register the frames, the time you will need for processing etc. A single pixel of a 4k frame represents - over the width of 5.69 mm of a S8 frame - a dimension of about 1.4 microns on the film surface. That is a very tiny, quite demanding dimension in terms of mechanics and optics.

Coming back to archival storage: besides scan resolution another important thing often overlooked is the dynamic range of film stock. Not many digital cameras can cope with the dynamic range of color-reversal film, even in raw mode (scanning a negative is a different story). To capture all the detail (for example the increased grain in dark areas) you need to capture several different exposures and combine them appropriately. Until HDR-displays become widely available and HDR-formats standardized, you are probably going to store just the raw captures for the time being for later processing. By the way, if you are opting for HDR-scanning with consecutive scan passes, you are going to have some difficulty in aligning different exposure runs of high-resolution images precisely, possibly introducing artifacts along the way.

Coming back to another use case, in a way opposite to the archival case: digital distribution of private movie material. Here, you might even want to reduce the grain scanned by your system. Usually, this dramatically reduces the file size you are going to distribute. And it seems to be less annoying for some people to watch such material.

In addition, non-professional S8-material usually is also very unsteady/shaky. If you compensate this as well, again the file size of your material will decrease while casual viewers will enjoy the material more. Of course, you are deviating a lot from the original material, but it is often irrelevant to your viewers, which enjoy simply the old stuff shown in the material.

In summary, I think it will depend on the specific use-case how one is going to scan old film material.

Anyone else tried Xojo before? It’s an interesting choice but I have no experience with it. I wonder how it would work if we need to include external libraries for CV, etc…

One downside is that to be a contributor, I’d either have to share the software license key so that one person could work on it at a time or contributors would have to buy their own license. I suppose I could also buy more than one and lend one out to a contributor while retaining the other one for myself…

Pro: speed and simplicity for X-platform uniformity (as long as there are no gotchas)
Cons: barriers to access for contributors (learning curve, price)

One thing I’d point out is that Xojo is totally free if you’re not going to compile into a standalone application. This means that another developer could use the programming environment at no cost, including the realtime debugger (which fully runs the application, but within the Xojo environment). I believe project files can be shared back and forth seamlessly between paid and free versions of the application of the same version.

As a self-taught programmer, I find it much easier to create a GUI application in a tool like Xojo than to do it from scratch or in a more barebones environment. And the built-in cross platform compatibility is nice. We’ve built GUI front ends for some command line applications that we use regularly, and they run on Mac, Windows and Linux machines. I mean, I’m making it sound simpler than it is, but this is an environment that makes it pretty easy to do this kind of thing.

Sony’s IMX series have amazing low-light capabilities with minimal noise. We just upgraded the camera in our ScanStation to 6.5k and it uses a Sony IMX342 sensor. We do see some advantage to 2-flash HDR with really dense film (like Kodachrome 2 stops underexposed), but the dynamic range in single-flash mode is pretty amazing for footage that just a little dense.

HDR can be done very easily with an intermittent motion because you can use exposure time as the primary variable you change between shots. With constant motion, you can increase the exposure time a little (which is what the ScanStation does), but it’s a balance because too long an exposure and you get motion blur. Alternatively, the light could be cranked up for the second flash but that also introduces some of its own issues. But the basic algorithms for doing this kind of thing are pretty simple. I think OpenCV even includes example code for this.

We offer this as a secondary format along with the scan for a nominal extra charge. Most people opt for it, because it gets them something that looks good enough, and is viewable immediately. This is typically a scan direct to MP4 (AVC), viewable on most televisions and computers right off a USB thumb drive. But this is done in conjunction with a high res scan, from which you can pull printable images, or in the future you can make higher resolution viewing copies as consumer displays get more resolution (without having to scale up the image, which will soften it).

Honestly, I’d say about 80% of the estimates we send customers turn into scans - there are always folks who want cheap and dirty, and wind up going to Walmart. Many have come back a year or two later because the scans looked so bad. By far the most popular resolution for home movies is 4k.

I’ve honestly never heard anyone complain about the grain (which if it really bothers you, is easy to manage in software on fairly modest computers). It’s simply not an issue that most people care about. The bigger concern is usually the speed of the footage (because nobody used to scan to 18fps, they’d either pull up to 24 or scan at 24 or 29.97 and the motion would be too fast or jerky).

1 Like

… with a high res scan, from which you can pull printable images, or in the future you can make higher resolution viewing copies as consumer displays get more resolution…

Good point, you can always go that way, but not the other way around :slightly_smiling_face:

My comment was a bit of hyperbole - you’re right in that grain is the picture. My assertion is that there’s a common sense limit of the degree to which grain needs to be resolved for users that don’t have a Netflix certification to hit. The pleasing round-off of grains is part of the character of the film when inspected closely, but that matters less when there’s motion (and gate weave etc from imperfect cameras). I can imagine that Errol Morris’ production wanted to provide high-resolution scans simply to intercut well with their F65 footage, which was likely 4k or 8k (I’d know for sure if they hadn’t rented their camera package from a competing house in NYC :slight_smile: ). It doesn’t indicate that there is truly “5K” worth of useful information to be gained from the actual film though. Useful for that project, yes, useful to me, no. Required by Netflix, probably, but they consulted Prodicle and likely had a rep to talk to. You know as well as I do (even better I’d hazard) that with the right optics and mechanics you could create a way to scan it at 8k, but there are diminishing returns in doing so. Maybe it’d be better to talk about this in terms of lp/mm?

In regard to the movement within the gate on later films, I haven’t seen it in a degree to argue that it’s more of an issue than the fact that many home movies were shot hand-held. The S8 films I’ve run through my scanner have seemed to be about as steady as the 8mm films (wrt frame position related to sprocket), although they had the noticeable benefit of generally being shot later (on objectively finer grain stock). I’m sure there are examples that run contrary to this, but this has been my experience so far.

Sure, storage is cheap, but not all Kinograph project users would come from the kind of background (and likely means) that either you or I do - for example, my current pile of film at home is going to occupy 30+ TB when it is scanned, and that’s before I get to actually do any work on the scanned files. I could buy 30TB of storage today but buying another 30+ TB for backup and then updating/upgrading that storage every other year or so for a home user is seriously pricey and time consuming.