Robert Houllahan

I always do an overscan for registration tests and on the Scan Station or Xena which have machine vision optical perf registration so the scan will be rock solid locked to the perfs.

Best practice IMO.

There are many ingredients to a process and one tank may be exhausted i.e. bleach fix which needs to be de-silvered and the developer is fine. It is very rare that a whole machine's full chemistry will be all dumped and replaced at once.

We run a sensiometric control before each run and check it on the densitometer and adjust chemistry as needed to get it into the proper spec to run.

Also there is a metered replenishment fed into the various chemistry tanks to keep it at spec for the run.

In theory and in practice a sensiometric control strip run at the beginning of the run and one at the end will be practically the same numbers on the densitometer and day to day.

On 8/7/2021 at 2:17 AM, Tyler Purcell said:

The registration on our scanner is very good, it's better than a Spirit.

Our solution to make it slightly better is develop a gate that holds the film firm in place with a spring loaded rail. 

I have a spare gate to do this with, just need someone to machine the parts. 

The Spirit 2K/4K has a spring loaded rail and a ceramic guide for the side closest to the film deck. Registration is fairly good for a telecine and also depends on what features are enabled on the machine.

It is not a mechanical or machine vision registered system and without one of those features in 2021 a scanner is relegated to being a dailies machine which will require allot of un-required work in post.

4 hours ago, Tyler Purcell said:

Got ya, so it's just AI machine learning software that reads the perf. Seems pretty straight forward to me. 

I was trying to come up with a solution that doesn't require the perf to be in frame. 

I would not exactly call it an AI but it has a certain area that it expects the perforations to be in and then looks for a specific shape and the contrast of it.

If you want to do really well stabilized film scanning without machine vision perf stabilization you will need a far better mechanical film transport than the ones on any of these new scanners. You can trigger the lamp/camera with a number of perf detector technologies from capacitive, laser or the optical encoder on the capstan.

A ScanStation can scan at 30fps or 15fps in HDR mode, the Arriscan in HDR is about 1fps. maybe slower, I am not sure about the speed on the new Arriscan XT but it does have a full immersion liquid gate option.

On 8/4/2021 at 1:29 PM, Tyler Purcell said:

Doesn't the Scan Station have an external registration capturing element, which it uses to real-time (via GPU) fix the files before dropping them to the drive? IE; it doesn't need to have the perfs in frame to register? 

The Scan Station has to be able to see the perforations with the imaging sensor to register the perfs and stabilize the film, there is no external sensor or mechanism. The transport is driven by a servo controlled capstan which has a optical encoder wheel attached to it, the film is driven at constant speed and then the machine vision perf stabilization does the rest.

6 hours ago, David Mullen ASC said:

Not sure if someone mentioned that a scanner doesn't need to use a Bayer color filter array and then demosaic the image, it can create RGB color in three passes for better quality & resolution per channel.

A number of currently available scanners do and with the newest sensors, it is just a hit in speed and there is high demand for quick turnaround times on film scanning.

The theory is that with enough pixels a CFA Sensor can make the equivalent of a real RGB sequential scan and I think that is somewhat true especially with 2X and 3X resolution oversampling. The issue is the CFA lenses are imperfect and there is allot of cross-talk between color channels compared to a sequential RGB scan so that crosstalk is worked out with math in the scanner software.

All that said there are customers we have that prefer the scans coming out of the new 6.5K sensor based Xena scanner we run to scans from the Spirit 4K or Scannity which are true RGB.

1 hour ago, Jack Jin said:

How much stops of dr does these sensors realistically have? Especially when compared to the DFT 4k and sdc2000. I know a lot of them can do multi pass hdr but I'm curious how these sensors stack up to the dft and sdc without the helpings of hdr. And how is the resolution from these bayer sensors compared to true rgb ccd sensors of the scanity and dft 4k?

I have a DFT Spirit 4K and the new 6.5K sensor based Xena I have has better performance in terms of detail noise performance and steadiness. We have client who prefer the scans from the 6.5K Xena to Scannity scans for OCN.

The SDC2000 Spirit is a very old and superceded telecine which does not even scan at 2K.

Looks like a optical sound reproducer for reading soundtracks from prints.

Not sure the make.

Rank-Cintel had absolutely nothing to do with the Thompson DFT Spirit, in fact Cintel and Thompson were competetors in the Telecine marketplace.

Cintel (then Rank bought them so it became Rank-Cintel) introduced the first really high quality Telecine in the 1970's and that is what this chassis is here, a MK3 Cintel Telecine.

Cintel used a system called "Flying Spot" that used a CRT to make the light for the scan, the Thompson Telecine used three linear CCDs and a hot lamp for the light source.

Cintel had several 4K scanners based on the Flying Spot tech at the end of their run, then BlackMagic Design bought Cintel.

The mechanical film deck and chassis on Cintel Telecine systems is second to none and is essentially a great platform for refitting as a modern scanner.

6 minutes ago, Perry Paolantonio said:

we are doing the perf detection in CPU. On a 14k image detecting the perfs and calculating the offset takes us less than 20ms. Granted we’re not scanning at 30fps. For that you’d need GPU almost certainly. Or you’d do it on the frame grabbers FPGA. 
 

Digital registration should improve with resolution. We’ve done some tests with the northlight (mech pin) and scanstation and found the scanstation was more stable. With faster scans they may have to simplify the file more than with a slower scanner, and that might lead to less accuracy. I’ll post some tests done on our scanner in a few weeks using opencv. 

Having been through this for years the GPU is generally where everyone has landed for machine vision based registration, the Xena allows you to select GPU or CPU and CPU is a significantly slower method that kills scan speed, and that is why LG does it in CUDA. I think it is the only way to get that job done at the speeds they can achieve with scan station, I don't think the frame grabber has the processing capability to do this in a fgpa there is just not enough processing at least for fast scanning.

As for the Northlaaaggg it is a line array and they drag the stage along for scanning the image, a not perfect mechanical arrangement, the Arriscan is an Area Scan sensor so the film and camera are held still for capture. Plus German fussiness.

Here is the "latest /greatest" sensor that companies are using in film scanners:

https://www.sony-semicon.co.jp/products/common/pdf/IMX342LLA_LQA_Flyer.pdf

6.5K 12-bit Global Shutter.

A machine vision camera and high speed frame grabber with this sensor will run about $7.5k-$8K

You cannot use a rolling shutter sensor for continuous motion film scanners like Xena ScanStation etc. it has to be Global.

For intermittent scanning (Step Frame-scan-StepFrame etc.) you can use a rolling shutter.

Here is a 14K sensor:

https://www.sony-semicon.co.jp/products/common/pdf/IMX411ALR_AQR_Flyer.pdf

Figure $16-20kK for that camera.

4 hours ago, Perry Paolantonio said:

It's all done on the image. frame registration is a trivially easy process using machine vision when you have something like perforations, which are well defined, to use as reference points. They might do this on the frame grabber in an FPGA, or they might do it in CPU or GPU, I have no idea.

Real-Time perf stabilization is done in GPU on all of these non mechanically pin registered machines.

BT at Co3 told me that they had done reg tests on the Scannity, the Scan Station the Spirit 4K and the Arriscan and that the Arri was the best registration of all of the systems they tested.

I was told that LG does all of this in CUDA language on the GPUs thus the speed, on our SSP the Scanner app is actually a 32bit one.

Xena uses OpenGL and Quadro GPUs so same result just slower than the LG machines which are sold as the fastest as a main sales point.

Most of the computer and imaging parts are off the shelf for these new machines with some customer hardware and then mostly software to glue the pieces together and make them all work.

The Xena works like the LG machine(s) for a color sensor it is 2-flash for a mono sensor it is 7 flash, so 2-flash per color plus IR.

We see people shooting print stock for the look and cost, it has a unique look and color.

I would suggest running it in a Filmo or a similar robust camera in case of a jam.

Also the "HDR" works by rewinding the film and doing a second pass and not by double flash while scanning like the Scan Station.

It is what it is.

Pretty poor for 16mm.

Effective Resolutions

3840 x 2880 - Super 35

3390 x 2864 - Standard 35

3390 x 2465 - Anamorphic 35

1903 x 1143 - Super 16

1581 x 1154 - Standard 16

1 hour ago, Phil Rhodes said:

 

These damn things are an exercise in historic computer research.

I was able to get my Data Spirit migrated to SUSE11 and PhantomII software which is much more modern.

Also while looking at these machines I think it is good to remember that it was not that long ago that none of this fancy digital high res sensor and fast gpu stuff existed.

Realtime 2K and 7-8fps 4K true RGB was pretty impressive when these were new in 2005ish on.

And post houses and labs which bought these scanners typically made that investment back in a year.

Spirit Data (2K 4K) machines are complex to setup, most run BONES on SUSE10 Linux and require very specific FC storage.

Spirit machines with the SDI video outs can be run standalone with the GCP controller or connected to a DaVinci 2K.

Spirit 2K 4K are usually 3-Phase power although they can be configured to run on 220v.

Not really machines for the uninitiated or those who think they are going to be an easy machine to run.

I think what you are seeing here (according to people who know) is a bunch of parts machines and random stuff, there are older DaVinci 888 panels and cards from various Spirits etc. I don't think these are running machines.

The Spirit 2K/4K series is a very reliable competent scanner but you do need a facility and parts machines to run them, good for a lab or post house definitely not for an individual in their house.

I have two at Cinelab and they are workhorse scanners and live along side our Scan Station and Xena machines.

List price on a new Spirit 4K in 2008 was about $2M

6 hours ago, Phil Rhodes said:

The sheer amount of electronics in those things is overwhelming - dozens of boards. Handling HD with all the colour processing in the 90s could reasonably demand that but I'm surprised it had to be so staggeringly overbuilt by 2008. I get the impression they were clinging to the original Spirit architecture and just beefing it up for 4K.

No not really at all, the two machines I have are quite a bit simpler in terms of number of boards in them, in fact the whole right lower section of the machine is empty on the last models. The basic film transport is similar to the rest of the Spirit classic and Quadra etc. machines as is the lamp and how the CCD arrays are placed relative to the gate but the overall electronics got simpler and more robust.

They are great machines if your a lab or big post house and there are some definite advantages for some material compared to some of the newest generation of scanners.

They are excellent scanners and were $2M in 2008 and many high end shows still use the Spirit 4K for scans.

It is a very reliable true RGB scanner but really needs a proper space to live in an engineering support including having a parts machine on hand.

In a ECN processor the step after the prebath is a rotary sponge which has a set of water jets, I would suggest building something with a rotary backing removal tool and water to carry the backing away.

Nice piece for a museum.