I don't mind the dirt. Can we see what the "rainbow" cleanup looks like?

Joel said:

If you have any big stuff to fix that doesn't get cleaned up with automated processes, I'd be happy to take a crack at it with AE rotoscoping (I outlined some of this in the x9 thread, not realizing that you'd moved on from that already :) )

The automated process is working rather well, actually. I'd like to see how far I can get with it before resorting to manual touch-ups.

My current problem is that it removes most of the starfields, so my spouse and I will have to sit down for a few hours to get the [begin, end] frame numbers for each scene containing stars in order to skip over them. Fortunately, ANH doesn't have many of these scenes.

I would estimate that 95% of the defects in the SWE Technidisc are removed with this automated process. I'd like to scrub through the remaining 5% manually, but it's 174,258 frames to look at. Can you (and others?!) help me find these defects? If so, I can provide you with my working copy.

Asaki said:

I don't mind the dirt. Can we see what the "rainbow" cleanup looks like?

Sure: Sample 5 Sample 6!

For Sample 5: Look at the ridges of Darth Vader's "nose".

For Sample 6: Look at the ridges of both ships (and elsewhere).

Wow! You've done some good clean-up on this so far and it sounds like you're only getting started. Filtering out the rainbowing was definitely a good choice; it's one of my bigger problems with the Technidisc. I'm looking forward to seeing what else you can do.

Now then, one problem that msycamore frequently brings up with his own transfer is the poor white balance as shown here. Are you planning on addressing that specifically?

bkev said:

Filtering out the rainbowing was definitely a good choice; it's one of my bigger problems with the Technidisc.

Yeah, I agree, it's very ugly but I ultimately decided to left it in, as the side effects of such filtering are often worse than the original problem. It's no match to get rid of, but what you often end up with is smearing, ghosting, objects that should have colors gets desaturated - seen on lights and lasers etc. I have not yet found a good derainbowing filter that not introduce artifacts of some sort. Fast moving scenes are the toughest.

msycamore said:

bkev said:

Filtering out the rainbowing was definitely a good choice; it's one of my bigger problems with the Technidisc.

Yeah, I agree, it's very ugly but I ultimately decided to left it in, as the side effects of such filtering are often worse than the original problem. It's no match to get rid of, but what you often end up with is smearing, ghosting, objects that should have colors gets desaturated - seen on lights and lasers etc. I have not yet found a good derainbowing filter that not introduce artifacts of some sort. Fast moving scenes are the toughest.

LUTDeRainbow works well. I plan to ColourLike against another source anyway.

Do you know of any filters for temporal color stabilization? Alternatively, is it commonly called something different?

bkev said:

Wow! You've done some good clean-up on this so far and it sounds like you're only getting started. Filtering out the rainbowing was definitely a good choice; it's one of my bigger problems with the Technidisc. I'm looking forward to seeing what else you can do.

Now then, one problem that msycamore frequently brings up with his own transfer is the poor white balance as shown here. Are you planning on addressing that specifically?

Thanks! I'm still experimenting, so this project can go many different ways.

I plan to address color-related issues later, as thinking about them gives me a bad headache. Ideally, I'd like to address each issue specifically. What changes to the white balance look best to you?

frank678 said:

What is/how does ColourLike work?

Click Here

After experimenting with ColourLike() using the Technidisc and the GOUT, I've decided that it doesn't work well enough; e.g., GOUT w/ Technidisc colors has weird patches of what are best described as murky rainbows.

frank678 said:

AntcuFaalb said:

frank678 said:

What is/how does ColourLike work?

Click Here

After experimenting with ColourLike() using the Technidisc and the GOUT, I've decided that it doesn't work well enough; e.g., GOUT w/ Technidisc colors has weird patches of what are best described as murky rainbows.

Do you have to have 2 complete films to do this. Can you use the values of a still and apply it to a whole scene?

Also it's possible myscamore might be willing to recapture the Technidisc if you can see a way forward with it that you can't get now, etc. I'm interested to see what you arrive at anyway whatever version you work on, some of these versions look so bad in places practically anything is an improvement!

1. Yes, you can use the values of a still for a whole scene, but I haven't tried it yet.
2. I've spoken with myscamore about this, actually. Due to some technology-related issues, he's unable to recapture the Technidisc at this time.

On an unrelated note: Now that I have three different sources of the JSC, I'm going to start experimenting with a new approach this evening; that is, I'm going to try using TooT() (two-out-of-three). (The three different sources are: V8, X9, and Dark/Sega.)

I'm not going to recapture it, because the results wouldn't be any different with the player or combfilter available to me, but like I've said before I'm going to upload a 2nd much improved version from my original capture as I know it can look so much better than what the current DVD does.

AntcuFaalb said:

Sure: Sample 5 Sample 6!

Not bad. It does miss a few spots, but it looks a lot nicer than it did before.

I have no idea what kind of negative effects from the process I would be looking for, though =)

AntcuFaalb said:

I plan to ColourLike against another source anyway.

 I don't know what you would mix it with...the GOUT has nothing to gain from...maybe you could get a tiny sharpness boost from adding the Technidisc's color with the JSC, but that seems like a lot of work for something that will (hopefully) be obsolete before the year is over.

Update: The TooT() idea won't work with the three different JSC captures, as they're IVTC'd differently and therefore won't sync-up.

I've been working with the Mallwalker-IVTC'd X9 and my results have been, surprisingly, much better than the samples I've already posted. The combination of a modified RemoveDirtMC and DFMDeRainbow really hits the sweet-spot.

I'm shooting for a release in 1-2 weeks. I'll post some samples tonight.

frank678 said:

Another forthcoming new release. good stuff.

PS78 might get in the way :-)