@drzoidberg33 Credit Detection is working OK for finding when the credits begin, and I see the “Skip Credits” button (at least on non-Android clients). Hitting that button ends playback since there is only the one credit marker in the xml file, marked ‘Final.’ However, in almost all episodes there is a “coming later this season” mid-credits scene that isn’t detected. Is this something the algorithm can be tweaked to catch?
Running Windows server, beta release 1.31.1.6733
No. We are not looking for scenes, we are looking for the credits. There isn’t a way to identify a “Coming next” or “Coming soon” section. The same problem goes for the beginning for “Last time on…”.
How is such a thing different than other mid-credits entities, like those in MCU movies? I thought the algorithm was supposed to identify (for lack of a better phrase) video elements in the middle of credits that are clearly not credits.
No, we look for the credits (i.e words on the screen). If these stop before the end of the file, we mark the end of the current credits. We then continue to look for more credits and mark these if found. The extras scenes are a by-product of this.
Edit - To be clear, if this “up-coming” scene is between credits, the credits should be identified as 2 separate sets of credits and the scene should be viewable. If this isn’t working, there may be an issue with detecting these credits. What show are you specifically referring to so I can check?
This is what I was referring to. I mentioned the TV show in my original post: ‘Your Honor.’ But going back to look at it, the scene in question is shown in a window with credits continuing to scroll alongside that window. The TV series ‘Shameless’ has the same thing going on. I realize that may be a special case since the credits are still playing, but an interesting one nonetheless.
The very last episodes of the TV shows ‘Andor’ and ‘WandaVision’ each have post-credit scenes. In those cases the scenes are literally at the very end after all the credits have played, and the detection algorithm isn’t picking them up.
I decided to see if that sort of thing was happening for movies, too, so I started looking at some of the MCU movies. Like those two TV show episodes, the movie ‘The Avengers’ (2012) has a scene at the very end after all the credits, which is not detected (there is another mid-credits scene that is detected OK). I’m sure other MCU movies have the same thing going on.
I’m merely pointing these out because it seems like you’ve been soliciting examples with the goal of improving the detection algorithm over time. Is there a better way to contribute this information?
Can you provide the XML info for any files you think are detecting the credits incorrectly?
Here are some examples I’ve found (mostly TV show episodes, with one movie):
Shameless (US) S11E01 - initial credits not marked; mid-credits scene not marked; final credits marked after mid-credits scene
Shameless (US) S11E01.txt (6.1 KB)
Your Honor (US) S02E04 - initial credits marked; mid-credits scene not marked
Your Honor (US) S02E04.txt (9.1 KB)
Your Honor (US) S02E05 - initial credits not marked; mid-credits scene not marked; final credits marked after mid-credits scene
Your Honor (US) S02E05.txt (6.0 KB)
Andor S01E12 - initial credits marked; post-credits scene not marked
Your Honor (US) S02E04.txt (9.1 KB)
WandaVision S01E09 - initial credits marked; post-credits scene not marked
Your Honor (US) S02E04.txt (9.1 KB)
Guardians of the Galaxy Vol 2 - initial credits not marked; first mid-credits scene not marked; other mid- and post-credits scenes marked
Guardians of the Galaxy Vol 2.txt (95.6 KB)
I tested all of these using the Windows desktop app. When I say something is not marked, I mean no "Skip Credits’ button is shown where expected, or when the button is shown it skips past subsequent mid- or post-credits scene(s).
I previously mentioned that the final post-credits scene in The Avengers was not marked. In testing this again, the player sometimes skips over it when the final Skip Credits button is shown, and sometimes it properly skips to that scene. So I think the scenes are marked correctly, but the player was behaving strangely.
Let me know if I can provide more information.
These are incorrect terminology and confusing, please do not refer to them this way. We do not mark the scenes, only the credits. Please indicate what credits are missed or if it’s too aggressive.
i.e. For “Your Honor s02e04” I see that we marked credits that don’t go to the end. If there are more credits after this, then “final credits not marked” would be the appropriate way to report this problem.
For “Shameless s11e01” it would be “earlier credits not marked”.
the player sometimes skips over it
If the player is not following the credits, that is a player issue. Please report these for the appropriate player along with the logs for that client. I’d like to keep this thread clear for the actual detection issues.
Got it. Do you need me to re-phrase what I sent, or do you get the gist?
I get it for now. Probably going to need some diagnostics from you later on to investigate. Checking if I can reproduce some of these first.
For WandaVision, your file is longer than I would have expected. Yours is about 54 minutes long. According to TheTVDB and TheMovieDB, that episode is 51 minutes. This may be contributing to the mis-detection.
Can you also check the “marker source” setting for credit detection?
Why would the runtime reported by TVDB or TheMovieDB effect credit detection? My understanding was that credit detection analyzed the video, basically looking for on screen text. And if the analysis does include looking online for that information, wouldn’t that be forcing problems with Extended Editions, NTSC vs PAL, etc?
I assume that Plex employees will test using what might be “standard” versions of each episode to see if the episode’s credits are found and marked. If THEIR copy has the credits marked, they can decide that their logic works for that episode. If your copy has 4 extra minutes of unknown fluff, maybe that interferes with the logic that they tested against, so the results are a surprise to them.
FWIW, my Wandavision S01E09 is 46 minutes long. The initial credits (slow credits, a flashy background but only the BIG names get credits one at a time) is missed. The first (and only) credits marker shows up after the first mid-credits scene, when the full screen white-on-black credits appear. The credits marker ends in time to show the final post-credits scene.
Wandavision S01E09.txt (10.0 KB)
Edit: I had to manually scan this episode myself, as apparently my server has not gotten to this show in its maintenance yet.
It is tracked based on a hash for the file (no we don’t track where the hash came from). A possibility is if the info for a specific hash is wrong due to an earlier version of the algorithm, and it gets reused. That’s why I’d like the OP to check if the cloud data might be being used. If not, then it’s just a flaw in the detection. If it is, then changing the setting to local only and a manual rescan might fix it.
I just discovered the xml files I sent were messed up. I somehow captured the wrong xml data for some of them (The WandaVision and Your Honor E04 xml files were actually for Andor). Not sure how I did that. 
Notwithstanding that screw-up, my WandaVision E09 is 46:51, well shy of the 51 reported in TVDB. My Andor E12 is 54:11, still shy of the 57 in TVDB. These videos have been processed to eliminate non-show specific content (e.g., streaming service ads, or whatever). I’m guessing that accounts for the differences.
My marker source was “both, try online first.” Since you asked about that, I switched it to “local detection only”, re-analyzed both the Andor and WandaVision episodes, and Skip Credits then worked properly - i.e., went directly to their respective post-credits scenes. Oddly, the re-analyzed xml files are identical to the originals, so go figure.
Doing the same for Your Honor had no effect. Doing the same for Guardians also had no effect.
EDIT: I wrote this after a bit more playing around, during which you added a few more thoughts to the thread. As you can see, the perceived time difference for WV was due to my error in providing the wrong xml. But still, switching the source to local and rescanning fixed it. But I don’t know why or how, since the xml data is the same before & after.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.
