InPaint Delogo - hardsub ripping - new user help please? :)

[VoodooFX]

So that's all done. The video now starts playing, except that it looks like it's 2 bit quality. (Probably supposed to.)

Now find HD video with subtitles as Hypocrites1915 doesn't have any.
It's black and white mask, a good mask is when you see only subtitles but don't see any other things.

Is the following correct?

No.
Extraction is activated with "Extract=1" parameter.

[Spiny Norman]

Now find HD video with subtitles as Hypocrites1915 doesn't have any.
It's black and white mask, a good mask is when you see only subtitles but don't see any other things.

It has title cards (intertitles), and automated extraction could really speed things up for those as well.

Extraction is activated with "Extract=1" parameter.

If you'd use a script for this example, what would it look like?

[Spiny Norman]

This is what I've got so far:

LWLibavVideoSource("C:\Swtools\Hypocrites1915.mp4" )
InpaintDelogo(Loc="60,80,360,180", Show=4, DynMask=4, DynTune=160, DynMask4H=70, Extract=1, ImgDir="C:\Swtools\")

That at least creates a file ~SubsMask2Img_FrameDB_20240920_125431_764.DB and a lot of image files. Some are false positives but others are promising, like the attachment.

How could I improve this further? I can also find a proper subtitle example (in colour) if that helps.

[VoodooFX]

Hypocrites1915 doesn't have any subtitles, learn it with normal video.

[Spiny Norman]

Hypocrites1915 doesn't have any subtitles, learn it with normal video.

That's what you said before, but I just posted a screenshot of a title card, so doesn't that prove that it's not impossible? Silent movies are a niche but they need OCR and translation too.

Anyway, it doesn't matter because there's plenty of other material. Something in colour then. Let's say this video: https://www.youtube.com/watch?v=rS-29jqkDUo
The resolution isn't terribly high, but to be honest that's really where a good script can make a difference. HD hard coded subs are crystal clear and even the GUI solution can get those in accurately.

[Spiny Norman]

Hopefully you are still willing to help with this example. It is exactly in this resolution and quality (sometimes even worse, but not often better) that I have many videos that would be so handy to OCR the subtitles from. In other words, if I can get this to work well, then it could really be extremely useful.

The results that I get are very promising but seem halfway there. This is the full script:

LWLibavVideoSource("C:\Swtools\Backanterna.mp4")
InpaintDelogo(Loc="94,368,280,86", Show=4, DynMask=4, DynTune=160, DynMask4H=70, Extract=1, ImgDir="C:\Swtools\")

I get for example this screenshot:

[VoodooFX]

You are just copy/pasting same command for different videos, do you think it somehow magically will work for everything?
Help you with reading the manual or what?

Told you already, start doing things with "HD hard coded subs are crystal clear" videos, when you have crystal clear understanding what you are doing with them then you can go for the difficult videos.

Now you are asking help to drive a car when you can't even enter it...

[Spiny Norman]

Now you are asking help to drive a car when you can't even enter it...

Quite so; I used that analogy myself when discribing this conversation to others.

Because it's also like the driving instructor won't demonstrate some essentials just by showing what it should look like. You praised this method as much better by all standards than whatever GUI there currently was. I want to see it in action on mediocre videos, because that's where a superior technology can make all the difference. I'm not sure if that's such a bad test case (just like I'm not sure that it won't work on silent movie title cards).

You could say for example, "my script for that video would look like this, because of these particulars (colours, shapes, background etc.)".

Explaining by example (problem- or case-based to to speak) could have saved you a lot of time as well (e.g. now you had to tip me to use LWLibavVideoSource instead of AVIFileSource).

Or, please point me more precisely to where that information is found, then I'll thank you, and start reading. If it's there, then I missed it. (Do you mean the Doom9 forum thread? That's 50 pages and a lot of it is concerned with removing logos.)

[VoodooFX]

You had plenty of demonstrations 3 years ago, where did that led?

Explaining by example could have saved you a lot of time as well (e.g. now you had to tip me to use LWLibavVideoSource instead of AVIFileSource).

Not doing your work saved me plenty of time already. Keep refusing to follow my instructions and it will save me even more time.
Btw, you was told about LWLibavVideoSource 3 year ago, but you was too lazy to read few pages back...

Or, please point me more precisely to where that information is found, then I'll thank you, and start reading. If it's there, then I missed it. (Do you mean the Doom9 forum thread? That's 50 pages and a lot of it is concerned with removing logos.)

The manual is at the top of the script, read about "DynMask" stuff, reading the forums may help too.

... a superior technology...

Your knowledge on the topic is lagging a lot, today's superior technology is AI based.

[Spiny Norman]

Your knowledge on the topic is lagging a lot, today's superior technology is AI based.

Yes, I imagine so, that's something to keep an eye on. I don't know of a practical option right now; who knows, perhaps in a year or so youtube will OCR everything perfectly - but then again, this particular question may not be top priority.

You had plenty of demonstrations 3 years ago, where did that lead?

Three years ago you claimed it blew other options out of the water, and I believed that, because you shared a very convincing result (one, not plenty, although it was enough). But not the method (i.e. the script).

The manual is at the top of the script, read about "DynMask" stuff, reading the forums may help too.

So presumably, when you extracted the example that I just mentioned from an arte broadcast, you did something with DynMask=5 and something with DynColor with a hexadecimal, because they use yellow letters? Can you narrow that down a bit please?

Yes, the LWLibavVideoSource command was mentioned earlier, but that doesn't disprove my point that examples (case-studies) are a great way to spread information and an efficient way to teach stuff to other people. Especially if someone asks for it specifically.

Wouldn't you see it as a triumph if you were able to teach this to someone who started knowing NOTHING about scripted editing? (I could also move the whole inquiry to the Doom9 forum, but the matter wouldn't really change; the main difference would be that other people might chime in.)

[VoodooFX]

Yes, I imagine so, that's something to keep an eye on. I don't know of a practical option right now; who knows...

Yeah, who can click that google search button...

...you shared a very convincing result (one, not plenty, although it was enough).

Not one... here is a quote from your second or third demo request:
"Is there any chance, that you could try what you get in this file, if I PROMISE it's the last time ever...? "

So presumably, when you extracted the example that I just mentioned from an arte broadcast, you did something with DynMask=5 and something with DynColor with a hexadecimal, because they use yellow letters? Can you narrow that down a bit please?

Presumably, yes. Narrow down what?

Yes, the LWLibavVideoSource command was mentioned earlier, but that doesn't disprove my point that examples (case-studies) are a great way to spread information and an efficient way to teach stuff to other people.

You already reposted one example, how much did that teach you? Search forums if you need more script examples.

Wouldn't you see it as a triumph if you were able to teach this to someone who started knowing NOTHING about scripted editing?

Presumably, yes, maybe, no.

[Spiny Norman]

Not one... here is a quote from your second or third demo request:
"Is there any chance, that you could try what you get in this file, if I PROMISE it's the last time ever...? "

The only one I recall was the German televised tragedy from Arte. (Wasn't that SD, by the way?)
Yes, I did ask for another - as you know, that was to try and help people out in that community. And you did not do it.

Now don't get me wrong: that is something that I can understand. It's in the "don't give a man a fish, but teach a man how to fish". You didn't want to do endless extracting on demand, preferring that people learn it themselves. So here we are.

Anyway, all this doesn't seem very relevant to me. It doesn't disprove what I said, and doesn't answer my questions. What would your script look like for any of the video examples?

It's very simple, providing that you actually want to help other people. If you HAVE the answers (or googled a better way), then you can prove how smart you are. And also, then it shouldn't take you long to give them.

I'm not asking for your tax returns! (But it's beginning to feel that way!)

[VoodooFX]

And you did not do it.

Any proof of that?

What would your script look like for any of the video examples?

You already have posted script examples yourself:
https://forum.videohelp.com/threads/404208-InPaint-Delogo-hardsub-ripping-new-user-hel...e3#post2751182
https://forum.videohelp.com/threads/404208-InPaint-Delogo-hardsub-ripping-new-user-hel...e3#post2751672

It's very simple, providing that you actually want to help other people. If you HAVE the answers (or googled a better way), then you can prove how smart you are. And also, then it shouldn't take you long to give them.

I'm not asking for your tax returns! (But it's beginning to feel that way!)

You got the answers, not sure what other answers you want.