Retro-Universal Scanner in Action!

[Tscan]

Here's a quick look at the new Retro-Universal in action! Just arrived on Weds and been testing it out with some fresh home movies shot on S8 50D. This unit does really well adding density to the negative during the scan if desired. I'll try to post a sample by the end of the week and some 16mm soon as well too.
https://vimeo.com/139775288

[MovieStuff]

Hi! Glad it's working well for you. Always nice to see our units in operation. However, you don't have it threaded properly. As noted in the instructions, the source reel must be locked down for proper back tension and, for 50 foot rolls, the film must bypass the first roller and thread directly under the film drum. This unit is different than your previous Retro-8 in that the back tension is created by a combination of the natural drag of the rewind motor gearing as well as the lower film drum. If you don't lock the reel to the feed spindle, then the film will be loose like we see in the video you posted. This can lead to bad registration, shifting focus and possible scratches because the film will rake across the top of the sensor block channel due to a lack of tension. If you are using reels larger than 50 feet, then you should feed the film over the first roller and then under the lower film drum. But, if using only 50 foot rolls, you need to bypass the first roller. Again, there are photos in the instructions that illustrate the proper threading for small and large reels.

Roger

[Tscan]

Thanks Roger, still finding by bearings a little but I see what you mean. The image quality still came out superb.

[Will2]

Great looking machine!

Did you just get the 8mm option or did you get 16mm as well? Interested in how easy of a change that is.

[milesandjules]

looking awesome.

[Tscan]

Great looking machine!

Did you just get the 8mm option or did you get 16mm as well? Interested in how easy of a change that is.

I did order the 16mm gate as well, it's getting delivered this week. I've been playing with S8 negative for now and can say this unit gives you a lot more exposure flexibility with negative film which is key. I'm really excited to see how 16mm looks.

[Will2]

...this unit gives you a lot more exposure flexibility with negative film which is key. I'm really excited to see how 16mm looks.

Roger,

I don't know if this makes sense, but what if the software could take three different exposures or even two then combine like HDR? Might not give much of an advantage but if film can capture 13 stops, does the camera capture that with one pass? I know it would take longer, possibly much longer but wouldn't it give an amazing result in the end? Just curious.

[Tscan]

...this unit gives you a lot more exposure flexibility with negative film which is key. I'm really excited to see how 16mm looks.

Roger,

I don't know if this makes sense, but what if the software could take three different exposures or even two then combine like HDR? Might not give much of an advantage but if film can capture 13 stops, does the camera capture that with one pass? I know it would take longer, possibly much longer but wouldn't it give an amazing result in the end? Just curious.

The scanner does 10 and/or 15fps, so scanning a reel a number of times with different exposure settings is not a problem. There was a guy on one of these forums that used a home made step motor rig and avisynth with some Kodachrome from the early 70's. The results were pretty awesome. It's something you could probably easily do without having to alter the Retro=Scan software, you would just need a script.

[Will2]

I'm most people wouldn't want to take the time on that since many operators are selling transfers to people on DVDs and speed is a big concern, but it would be interesting to be able to take two exposures per frame so they are both taken with the film at the exact same position to make combining easier. Maybe create two separate files/folders of tiff files and use a Photoshop HDR plugin or something to combine. Just an idea for us image nerds.

[carllooper]

Doing HDR capture, through multi-capture, gives incredible results.

While the range of modern film is nominally 13 stops (which is incredible enough) you can actually get even more than this out of a negative. Normally the toe and shoulder of the characteristic curve takes a detour from the range in which DR is typically assessed, compressing shadow/highlights below the contrast threshold a sensor could normally distinguish. But using HDR in conjunction with some inverse curves with respect to the characteristic curve for a given stock, one can undo the compression of details that otherwise occurs in the toe/shoulder area, ie. extract a few more extra stops of information out of the negative.

And so reconstruct an even richer signal.

A very small problem you might have, if scanning the film a number of times, can be very small changes in frame position between one scan and another. If such changes are present it can reduce the sharpness of the composite result. So it's better, when possible, to do the multiple exposures with a frame sitting stationary in the gate.

But otherwise HDR can easily make up for any tiny losses in sharpness.

C

[supa8]

Interesting stuff. I was going to say the same thing about regristration. Surely, it would work better on a system where registration is dead-on steady via sprocket-base transport?
Even with HDR merging of frames, which would alone require huge amounts of processing, some artifacts would be bound to appear trying to superimpose two super8 images, whose frame position is slightly off, correct?
The camera would also play a big enough role, as only newer cameras will have little noise incurred in the transfer.
Is there not a risk you would double the amount of noise between 2 frames?

[carllooper]

Interesting stuff. I was going to say the same thing about regristration. Surely, it would work better on a system where registration is dead-on steady via sprocket-base transport?
Even with HDR merging of frames, which would alone require huge amounts of processing, some artifacts would be bound to appear trying to superimpose two super8 images, whose frame position is slightly off, correct?
The camera would also play a big enough role, as only newer cameras will have little noise incurred in the transfer.
Is there not a risk you would double the amount of noise between 2 frames?

You might blur the result a little but you won't magnify the noise. Quite the opposite: you'll reduce the noise.

You can test this in Photoshop. Import an image. Duplicate it twice. To each duplicate add some noise. Now blend the two noisy images together. In the result will be a reduction in noise compared to each considered on their own.

While at any given pixel the noise (the error) with respect to a signal (what the pixel should ideally be) will tend to possess a non-zero magnitude (ie. will tend to be a non-zero error), the set of all errors across an image will be as many times positive as it is negative (with respect to the signal: the image proper). It is this fact which guarantees that over time and/or space (in the image domain), errors due to noise will cancel each other out (will tend towards such), ie. a positive error plus a negative error equals a zero error.

s + e - e = s

C

[Tscan]

At 10fps I might worry about my system keeping up with a multi capture? The Retro-Scan uses sprocket hole detection with a variety of sensors which could be pretty consistent from one scan to the next, but Roger would have to answer that. One thing i noticed on the last multi-capture I saw was the grain was really tight as well.

[MovieStuff]

I'd say that running the same film through twice might work if you don't change the sensor knob between runs. It's hard to say. Clearly, this unit isn't pin registered and even the very best mechanical projector advance will not position the film in exactly the same place twice. Similar to a Rank, the Universal has a certain about of minor "float" to the image that isn't objectionable for normal viewing but whether that "float" is repeatable, I couldn't say because it wasn't designed for repeat registration. That said, running the unit at the slower 10fps speed would probably be better than the 15fps speed. We are working on a few specialty gates that would run very slowly, advance and stop, and allow for multiple exposures on a single frame before advancing again. These gates would also be good for use with micro 4/3rd cameras like the Panasonic GH4 and the such. We probably won't be releasing these gates until after the new new year as we are absolutely slammed with orders right now. I'm in the process of completing a special 28mm gate for the University Of Southern California Hugh Hefner Archive and also a unit for the Academy of Motion Picture Film Archive, so time is at a premium with the holidays approaching fast.

(shameless plug, sorry) For those that are interested, we are on our last day of our Labor Day 2-for-1 sale. Buy the unit with one gate and get a second gate for free!

Roger

[carllooper]

The noise being reduced in multi-capture is mainly reduction of sensor noise rather than film noise. Sensor noise differs on each capture, and on each pixel. So this is why multi-capture is able to reduce this noise.

Now noise occurs in film as well, which we call grain, and here it will be frozen into the image, so it can't be directly cancelled out through multi-capture. But we need to be careful here because there are two types of grain in film, one which is burnt into the film itself (native grain), and another which is a function of interference between any copy and it's original, called "grain aliasing". Indeed what we always see (observe or copy) in any image is not the native grain as such but this second kind. A kind of "fake" grain. It is an amplified version of the original grain. It is a function of observation. This second type of grain (unlike the first) can be reduced through multi-sampling but requires some additional work where the sensor is micro-shifted (sub-pixel shifted) between each capture, and then the digital copy sub-pixel shifted back again, prior to a merge. So in many respects it might be advantageous to capture the image at different positional offsets each time (randomly being better than otherwise), and then using digital registration to realign the captures with each other.

But even if one doesn't re-align the image there will be a reduction in grain because both types of grain are only frozen in terms of time. In terms of space, it will still vary from one point to the next across space, at all scales (down to the quantum physical). In other words the noise has an opportunity to cancel out through small changes in space: ie. misaligned errors mean positive errors will get an opportunity to line up with negative errors and cancel out. However this shift in space will reduce sharpness but not in any way that is necessarily objectionable. Just depends on how much.

Indeed if doing multi-capture on an image which is otherwise perfectly stationary with respect to the sensor, there will be no reduction in grain aliasing. The only way to decrease the grain aliasing in this case would be to increase the sensor's pixel count. This is in fact the best way to decrease grain aliasing. Increase pixel count, or otherwise do a film to film transfer and watch the result on film.

C