The Matrix [spoRv] *BD-25 RELEASED* — Page 5

CatBus said:

The question is: does it more-or-less grain-match the existing project?  Considering the final v2 may mix-and-match sources, it'd be nice if they blended seamlessly.

Oh, I was talking about doing it to my copy only. I'm a big fan of the open matte and I'd prefer to keep it as close to the original HDTVrip as possible.

I can post some samples and, if people like it, do a 720p release or something...

@CatBus: even if I prefer subtitles out of the picture, I think many people don't; so, if you want to provide adjusted subtitles for the V2, you are welcome!

@AntcuFaalb: can you explain better your technique?

PM sent!  I didn't fix the ugliness of the subtitles, just the positioning ;)  It makes sense that these are DVD upscales--that would explain a lot.

_,,,^..^,,,_ said:

@CatBus: even if I prefer subtitles out of the picture, I think many people don't; so, if you want to provide adjusted subtitles for the V2, you are welcome!

@AntcuFaalb: can you explain better your technique?

Sure. First, you download a 35mm grain plate like the one here: http://7dblue.wordpress.com/2012/03/07/flim-grain-for-everyone/

(Note: The following assumes that the 35mm grain plate clip and the target clip are the same resolution, e.g., 1080p.)

Then, convert it to a more friendly format, such as a Lagarith-compressed AVI.

Next, get it into Avisynth, like: grain = AVISource("C:\Grain35mm.avi")

Then, use Loop to make it the same length as the clip you're applying it to; e.g., grain = grain.Loop(1440) for 120-frame-long grain and a 2-hour-long 24fps clip.

Finally, apply it to your target clip; e.g., AVISource("C:\TheMatrix.avi").Overlay(grain, mode="softlight")

Thanks CatBus, really appreciated; they will be used when (if) the V2 will be released.

Delete some PMs, Andrea!

...ooops... done!

Here you are some new screenshots; top image is HDTV, second DVD, third ColourMatch (not tuned), bottom BD luma and BD ColourMatch'ed to DVD chroma (click for large version):

The fact is, when the sources have any clipping whites, the filter works flawlessly; I'm testing it with Halloween, and, even if it's not 100% color accurate, it *seems* that there is no artifacts. Obviously, if the clipping whites are present, you'll find something like the last image...

The HDTV cropping shifts wildly from scene to scene, so it's not possible to do a crop-it-all; it should be cropped scene by scene...

_,,,^..^,,,_ said:

The HDTV cropping shifts wildly from scene to scene, so it's not possible to do a crop-it-all; it should be cropped scene by scene...

Which HDTV cap are you using? The one on Usenet is rock-solid; that is, there's no shifting cropping like you mention here. It's 100% 16:9 (Super35 open matte) besides ~8 pixels on the left and right.

Sorry, I intended the cropped final frame, to match the BD... sometimes the top border should be cropped more than the bottom, the scene after the contrary is true... I thought to do a crop-once, to match the BD, but it is not the case...

By the way, the HDTV I'm talking about is a russian one, 1080i mpeg2 29.97fps hard telecined, divided in three .mpg files. And it seems that this one has not tiny black borders on the left, apart on few scenes (like the one in the last image in my previous post) where there is a 4-pixel border on the left.

_,,,^..^,,,_ said:

By the way, the HDTV I'm talking about is a russian one, 1080i mpeg2 29.97fps hard telecined, divided in three .mpg files. And it seems that this one has not tiny black borders on the left, apart on few scenes (like the one in the last image in my previous post) where there is a 4-pixel border on the left.

OK, it sounds like we have different HDTV caps.

AntcuFaalb said:

_,,,^..^,,,_ said:

By the way, the HDTV I'm talking about is a russian one, 1080i mpeg2 29.97fps hard telecined, divided in three .mpg files. And it seems that this one has not tiny black borders on the left, apart on few scenes (like the one in the last image in my previous post) where there is a 4-pixel border on the left.

OK, it sounds like we have different HDTV caps.

So you have a different HDTV version? Does yours have the missing ending credits?

ilovewaterslides said:

AntcuFaalb said:

_,,,^..^,,,_ said:

By the way, the HDTV I'm talking about is a russian one, 1080i mpeg2 29.97fps hard telecined, divided in three .mpg files. And it seems that this one has not tiny black borders on the left, apart on few scenes (like the one in the last image in my previous post) where there is a 4-pixel border on the left.

OK, it sounds like we have different HDTV caps.

So you have a different HDTV version? Does yours have the missing ending credits?

Interesting. No, it doesn't, but it also doesn't have all the wacky problems others have brought up.

drngr said:

The one on Usenet is just a binary combine of the three Russian .MPG files.

There are multiple issues with that HDTV capture, some of which were discussed in the thread that spawned this one.

http://originaltrilogy.com/forum/topic.cfm/The-Matrix-with-original-theatrical-color-timing/post/638963/#TopicPost638963

• At least a couple glitches in the recording, and the pulldown pattern changes at least once.
• There is horizontal aliasing. It looks like they took a 1440x1080 HDCAM master and resized it to 1920x1080 using point-resize. This can be fixed by downscaling and upscaling horizontally with a softer resizer. (I did that for the screenshot comparison)
• Recording stops during end credits.
• Being Super35, they cropped the sides off a bit to make the open matte transfer.
• Compression.

Thanks for this information!

I think I was able to fix the glitches with "force film" in DGIndex. It fixed four field-order transition errors.

I'll check for any combing using TIVTC later tonight.

My plan is to downscale to 720p, which should reduce some of the visible macroblocking. I haven't decided what to do about the aliasing.

drngr: I sent you a PM.

OK, the following image is one of the worst in the entire HDTV cap. The macroblocking is terrible, but I believe it's significantly harder to see when softlight-overlaid with an accurate 35mm grain plate.

http://screenshotcomparison.com/comparison/48037

The 35mm grain plate used here was generated from real 35mm film stock. IMO, it's too coarse. I'm working on getting a finer one.

drngr said:

Resizing to 720p should effectively blur the aliasing. The glitches themselves aren't fixed by correcting field order transitions.

Damn. Do you happen to have any frame-numbers I can check?

DoomBot said:

AntcuFaalb said:

OK, the following image is one of the worst in the entire HDTV cap. The macroblocking is terrible, but I believe it's significantly harder to see when softlight-overlaid with an accurate 35mm grain plate.

http://screenshotcomparison.com/comparison/48037

The 35mm grain plate used here was generated from real 35mm film stock. IMO, it's too coarse. I'm working on getting a finer one.

Nice job with the grain overlay, i like it.

Thanks. IMO, it does a really good job of hiding the macroblocking.

Andrea: It also does wonders at hiding noise in LD captures! :-)

The idea of actual film grain is great!

Curiosity: how a grain plate is captured? From actual empty 35mm cell frames?

_,,,^..^,,,_ said:

The idea of actual film grain is great!

Curiosity: how a grain plate is captured? From actual empty 35mm cell frames?

Honestly, I have no idea. :-(

drngr said:

AntcuFaalb said:

drngr said:

Resizing to 720p should effectively blur the aliasing. The glitches themselves aren't fixed by correcting field order transitions.

Damn. Do you happen to have any frame-numbers I can check?

My frame numbers are with Honor Pulldown Flags and then a manual IVTC attempt.

99465 (1:09:08). I also have a note saying "video glitches when Neo turns and tries to shoot agent that dodges" which I assume refers to 1:46:14, but I don't see anything wrong there at a casual glance now.

79962 is where one pattern break happens (at a scene change).

Would you be interested in working on IVTCing this together?

Two heads are better than one and all that...

drngr: OK, I busted-out some manual IVTC action. I used DGIndex to generate the d2v and chose "Honor Pulldown Flags".

It, again, found the field order transition errors and I told DGIndex to fix them. I then used the fixed d2v (rather than the d2v.bad) and did my manual IVTC.

I used IsCombedIVTC (from TIVTC) with ConditionalFilter to check for residual combing (indicating a pattern break) and told it to output a red-colored frame whenever it detects any.

I'm not seeing any red-colored frames with the fixed d2v, but I see lots with the d2v.bad.

Here's my short script: http://pastebin.com/5UCZcPq1

Please execute it and let me know if you get the same results.