Log in

View Full Version : Superior field-matching filter needed


Katie Boundary
10th June 2016, 22:36
I was dissecting some Transformers episodes today and noticed a tiny flaw in Uncomb: it occasionally spat out interlaced frames in places where there were no orphaned fields and therefore no reasonable excuse for why this would happen except crappy code. I dusted off the documentation for telecide, only to be reminded of why I stopped using telecide so many years ago. So now I'm looking for a field-matching filter that's more accurate than Uncomb, but doesn't have 500 tons of bloat and garbage attached to it like Decomb does. Another way of putting it would be that I'm looking for something that works just like Uncomb but without the annoying habit of generating obviously incorrect frames for no reason. Any suggestions?

hello_hello
10th June 2016, 23:02
Until someone's able to suggest a more configurable field matcher that needs no configuring, why not post a sample of the problem video? Just to eliminate user error.

johnmeyer
10th June 2016, 23:53
Any reason you are using Uncomb? I can't find any version that is newer than twelve years old. By contrast, IVTC has been updated dozens of times. I would recommend using TFM()/TDecimate().

Katie Boundary
10th June 2016, 23:54
Today, I shall humor Hello because it has been at least 4 hours since I've gotten into an argument with someone who couldn't understand basic English, and thus I am in a good mood.

post a sample of the problem video

In these two images, the combing shows up in the lines that make up the grids:

https://scontent-lax3-1.xx.fbcdn.net/v/t1.0-9/13418864_1699471563650632_5304436214019958884_n.jpg?oh=56c403c13277be4e2fd5f411ae6c232f&oe=58044672

https://scontent-lax3-1.xx.fbcdn.net/v/t1.0-9/13423944_1699471566983965_2833416500952847127_n.jpg?oh=254213d34b6b0e4ed0f36e3f08505af0&oe=580000C6

In this one, Ravage is interlaced:

https://scontent-lax3-1.xx.fbcdn.net/v/t1.0-9/13423804_1699471570317298_3591599239860417883_n.jpg?oh=81746cceb6d9154e17bf3bc71617f6a9&oe=58084619

Obvious interlacing is obvious:

https://scontent-lax3-1.xx.fbcdn.net/v/t1.0-9/13417642_1699471593650629_2849066176246791305_n.jpg?oh=9d326129745475d54230b79d679df28b&oe=57D741E7

Any reason you are using Uncomb?

It's fast, there's nothing to configure, and it doesn't try to do anything other than match fields.

I would recommend using TFM()/TDecimate().

I'll use TFM if I have to, but the documentation does a very poor job of explaining what all the different modes actually do, and I don't like the default post-processing setting.

johnmeyer
11th June 2016, 00:00
It's fast, there's nothing to configure, and it doesn't try to do anything other than match fields.And, apparently, it doesn't work.

Try TFM() and TDecimate() (they are part of IVTC).

I echo Hello's excellent point that I think he was making, namely that if you want to get good results, you will have to feed the plugin some settings. Every video is different, and that is exactly why most plugins have all sorts of settings. There are virtually no plugins of any type (that I know of) that work every time on all video, without doing some tweaking.

[edit]You changed your post. I replied to the wording you had in your original post, before it was changed (see quote above from your original wording).

poisondeathray
11th June 2016, 00:20
I'll use TFM if I have to, but the documentation does a very poor job of explaining what all the different modes actually do, and I don't like the default post-processing setting.

Then adjust it (You can adjust the coming detection threshold for example) or disable it ( TFM(pp=0) will disable it )

Katie Boundary
11th June 2016, 00:28
I've spent the past several minutes trying to track down a copy of TFM (which I used to have before the hard drive that it was on died). Unfortunately, it looks like the missouri.edu server is down. Does anyone have a backup or mirror?

And, apparently, it doesn't work.

Oh, it does... it just works very badly with this particular content.

poisondeathray
11th June 2016, 00:33
I've spent the past several minutes trying to track down a copy of TFM (which I used to have before the hard drive that it was on died). Unfortunately, it looks like the missouri.edu server is down. Does anyone have a backup or mirror?


TIVTCv105
http://www114.zippyshare.com/v/gdI0VTZl/file.html

Katie Boundary
11th June 2016, 00:40
Thanks! Now I have to download and install 7zip because some people are apparently too elite to use normal zip files like the rest of us peasants...

EDIT: Yes, the default mode and pp=0 yield precisely the results that I was hoping for. This will serve my needs very well. Thank you, John Meyer and Poison Death Ray!

poisondeathray
11th June 2016, 00:44
Thanks! Now I have to download and install 7zip because some people are apparently too elite to use normal zip files like the rest of us peasants...

ok fine pollute the internet :D

TIVTCv105.zip
http://www88.zippyshare.com/v/Ua3mYJtI/file.html

447KB vs. 277KB. It's a significant difference

hello_hello
11th June 2016, 00:47
Edit: I just realised. I must be off the ignore list. Oh the sheer joy of it...... :)

Today, I shall humor Hello because it has been at least 4 hours since I've gotten into an argument with someone who couldn't understand basic English, and thus I am in a good mood.

I'm in a good mood too, which is why I'll only point out you can't humour someone unless you're smarter than they are, although I realise you mightn't understand, given you're not able to distinguish between a sample and a screenshot.
Screenshots don't prove a lack of user error.

I'll use TFM if I have to, but the documentation does a very poor job of explaining what all the different modes actually do, and I don't like the default post-processing setting.

Do you prefer the combing? Maybe you should leave it as it is?
I'm sure there's a highly configurable field matcher that doesn't require configuration out there. Hopefully someone will know......

hello_hello
11th June 2016, 00:54
In these two images, the combing shows up in the lines that make up the grids

It looks more like a single field has been resized. Strange things happen to DVD content. It might be nothing to do with Uncomb. It could also be user error. It's hard to tell without a sample.

In this one, Ravage is interlaced

Combing under the credits. Mmmmmm..... mightn't be Uncomb's fault. It's impossible to tell without a sample.

Obvious interlacing is obvious:

It's obvious when the credits are over top of the video too. It mightn't mean anything, but screenshots don't tell the story. A sample would be nice. Wouldn't it be funny if it wasn't pulldown but it's interlaced? And you try another IVTC method that de-interlaces? And you like it?

Have you noticed every screenshot you posted shows combing under credits, which appear to have no combing, but it's impossible to even tell if they're moving, due to the continued confusion between a screenshot and a sample.

Katie Boundary
12th June 2016, 06:33
Sometimes I wonder why Hello even bothers posting in the threads that I start, knowing that he/she is on my ignore list.

hello_hello
12th June 2016, 06:58
Sometimes I wonder why Hello even bothers posting in the threads that I start, knowing that he/she is on my ignore list.

Because I think you're a liar and you do read my posts, given you reply to them when it suits you (http://forum.doom9.org/showthread.php?p=1769923#post1769923), then go back to pretending you're not reading them when it doesn't. Even when you're pretending to ignore them you still can't help responding to me, which is a bit of a give-away.
It's not just about you though. It's more about ensuring nobody stumbles upon your threads and assumes they should be taken seriously.

wonkey_monkey
12th June 2016, 13:18
Sometimes I wonder why Hello even bothers posting in the threads that I start, knowing that he/she is on my ignore list.

First of all, how would you know?

Secondly, whether you read his posts or not, he is doing others a service by pointing your numerous errors. You may never listen, but others will.