[vlc-devel] IVTC testing notes

[Juha Jeronen]

Hi all,

For those interested, here are some testing notes from my own tests.

The comments about the "new" filter refer to the "PATCH 5/4" version
just posted. All referred anime are US (R1) NTSC DVDs.


Stellvia still gives the new filter trouble, but with most other anime
DVDs it seems to work better now.

Suggestions for "standard" test material follow.

Difficult talking scenes (only a few pixels of motion in total, and that
at 8 or 12 fps):

- Utakata ep. 1, 11:05-12:05: works.
- The Third ep. 1, 11:55-13:00, 20:01-20:25 and 20:52-21:15. The second
mentioned part shows one interlaced frame when Iks talks (if you look
closely), but the rest work correctly now.

Difficult global motion (slow vertical pans, slow zooms, or a
combination of these):

- Marimite S4 ep. 1; 01:45-01:55 (in the OP!), 02:00-02:05, 02:10-02:14.
All work correctly now.
- Utakata ep. 1, 00:07-00:11, 03:35-03:45, 11:46-11:51, 13:45-13:48,
14:20-14:23. All work correctly now.

Cadence catching trouble (for testing the emergency mode):

- Utakata ep. 1, the dream scene right after the OP. Look at the debug
output. Pans, ok. The ocean waves - the cadence is not found and the
filter uses the emergency mode.
- Stellvia ep. 1, OP. Switches between soft and hard telecine very
often. Hard to IVTC correctly, good test material.

Other useful scenes for testing:

- The Third ep. 1, the space junk right at the beginning of the ep (just
after OP). This will show if the fieldwise motion threshold tunings are
too low ;)

- Azumanga ep. 1. Note that OP and ED are 30fps progressive (except some
texts, which are 60i). Ep itself is 24fps hard telecined. If the filter
switches modes correctly (i.e. if the detectors are good enough), it
shouldn't damage the OP/ED. See especially the slow vertical motion in
the OP at 00:13-00:21. Also, see 03:33-04:03 in the ep. Talking, then
some cuts, and finally a pan where the last frame gave trouble to the
previous version of the filter. Letting it play until 04:11, some
private development versions had trouble after the cut at 04:08. The
current one works fine with all this.

- Azumanga ep. 2, 12:30-12:33. The vending machines give some idea of
how large interlace scores can be for frames that are actually progressive.

- Kanon (2006) vol. 3, start of first ep (until OP). This will show
whether the subtitle flicker problem occurs.


Known artifacts with the "PATCH 5/4" version of the filter:

The Third ep. 1, 10:24-- when Honoka jumps up (at 10:38). I don't know
if this is because the field does not have a pair, or whether it's just
handled incorrectly. I would need to open this in a video editor to find
out for certain.

The Third ep. 1, 20:01-20:25 - one of Iks's lines. The rest of the scene
works. (This is the same one mentioned above in talking scenes.)

Stellvia ep.1, 5:05-- when Masato explains that he's doing his part as
the little brother. The same episode shows a few more artifacts, too.
Stellvia is hard to deal with, since it constantly switches between hard
and soft telecine.

Utakata OP, crossfades. This is probably due to fading between streams
with different cadences and can't be fixed in a simple manner.

And maybe some more, which I don't recall right now.


 -J