Theora encoder and frame drops/dupes

[rwill]

Hello,

I tried to encode some videos in Theora with the libtheora 1.2.0 encoder, either integrated in ffmpeg or the standalone example app, and noticed that whatever settings I use (target bitrate, soft target, target quality) it drops input frames and replaces these with duplicated frames when decoding. It seems the only way I can prevent this is to set the target bitrate to a value higher than Theora can reach when running at its lowest quantizer, so for FHD around 70-100Mbit.

Does anyone know if this is expected behaviour or am I missing something?

[LMP88959]

I vaguely recall running into this as well and realized frame dropping is part of the rate control code. Check out TH_RATECTL_DROP_FRAMES in the Theora code maybe and see where that gets exposed so you can disable that bit in TH_ENCCTL_SET_RATE_FLAGS.

[rwill]

Thanks, this works. Now VLC still only displays every 2nd or 3rd frame but Windows Media Player works correctly at least...

[kurkosdr]

Out of curiosity: Why are you encoding things with Theora? VP8 is much better and is royalty-free (Google paid an one-time fee to the patent holders that asserted patents against VP8 in order to make them go away).

Is there a specific hardware compatibility need that requires Theora?

[benwaggoner]

Bear in mind Theora is a slightly tweaked version of the VP3 codec, Ala five generations before even VP8. Three generations before the VP6 that was popular in Flash before H.264 replaced it.

On2 asked me to help licensing VP3 away to Terran Interactive circa 2000 because it was already uncompetitive by that point compared to their newer codecs. Rate control was particularly tragic.

I think I used VP3 for maybe one late 90's CD-ROM project.

I never used Theora for anything. It wasn't even competitive with MPEG-1, which also didn't have any IP issues, but with less compatibility and much less predictable file size.

[rwill]

Out of curiosity: Why are you encoding things with Theora?

Now thats.... a secret :)


*edit
By the way, this is misinformation:
VP8 is much better and is royalty-free (Google paid an one-time fee to the patent holders that asserted patents against VP8 in order to make them go away)

Google only made MPEG LA go away, not possible others.

[kurkosdr]

By the way, this is misinformation:


Google only made MPEG LA go away, not possible others.
MPEG LA is not a patent holder, it's a patent pool administrator, and all companies that were about to assert patents against VP8 were under the MPEG LA umbrella. As of this date, I'm not aware of any patent assertions against VP8.

FYI, no standard is guaranteed to be free of patent assertions, even the ancient MPEG-2 is covered by 2 "submarine patents" that surfaced in Malaysia (the last one in 2020). Similarly, unless all pre-TRIPS patents have been published in the US, there is the possibility of a "submarine patent" covering even MPEG-1 (and later) to surface tomorrow in the US. It's like the "fix bugs before adding new features" rule. There is no way to guarantee a piece of software is bug-free (not in practice anyway), what the rule actually means is "fix all known bugs before adding new features". Similarly, when we talk about "no patent assertions", we actually mean "no known patent assertions". And this will stay true until all "submarine patents" are published in all countries.

[kurkosdr]

I never used Theora for anything. It wasn't even competitive with MPEG-1, which also didn't have any IP issues, but with less compatibility and much less predictable file size.
A person from the FOSS camp told me that Theora was about the same level as MPEG-4 Part 2, but this was during the "codec wars" of HTML 5, aka when W3C tried to reach a decision on which video standard will be mandatory for the HTML 5 "video" tag (before W3C gave up entirely and left the decision to the implementers), so obviously that FOSS person had a motivation to embelish Theora's performance.

On the other hand, I find the idea of Theora falling behind MPEG-1 equally implausible (maybe initial implementations of the official encoder, which were notoriously badly-tuned, but not the latest version).

Generally, I would love to see a reliable, apples-to-apples (BD-Rate) and oranges-to-oranges (same encoding speed approximately) comparison of Theora versus the MPEG-4 Part 2 (using the Divx and Xvid encoder) and H.264 (using x264 and MainConcept), but since Theora saw very little adoption, it's often not included in benchmarks.

[rwill]

MPEG LA is not a patent holder, it's a patent pool administrator, and all companies that were about to assert patents against VP8 were under the MPEG LA umbrella. As of this date, I'm not aware of any patent assertions against VP8.

[snip]

Oh please, don't think I am an idiot.

Google paid of MPEG LA in March 2013, this here happened at around the same time:

IPR: https://datatracker.ietf.org/ipr/2035/

One of the still remaining articles which stayed online about the Nokia case: https://www.windowscentral.com/nokia-refuses-google-access-patents-vp8-codec-dream

[kurkosdr]

Oh please, don't think I am an idiot.

Google paid of MPEG LA in March 2013, this here happened at around the same time:

IPR: https://datatracker.ietf.org/ipr/2035/

One of the still remaining articles which stayed online about the Nokia case: https://www.windowscentral.com/nokia-refuses-google-access-patents-vp8-codec-dream
That was 13 years ago, the fact Nokia didn't win any court judgments means the patent assertions were unsuccessful.

[rwill]

That was 13 years ago, the fact Nokia didn't win any court judgments means the patent assertions were unsuccessful.

Ah kurkosdr, thats why I am so fond of you. You went to the above from this quote here in only 2 posts.

As of this date, I'm not aware of any patent assertions against VP8.


You always speak full of confidence but what does this confidence actually mean outside in the real world? Shall someone base his multi-million dollar business on such statements? Are you going to foot the bill if some IP holder comes after them?

[kurkosdr]

Ah kurkosdr, thats why I am so fond of you. You went to the above from this quote here in only 2 posts.


You always speak full of confidence but what does this confidence actually mean outside in the real world? Shall someone base his multi-million dollar business on such statements? Are you going to foot the bill if some IP holder comes after them?
Again, no codec is safe from patent assertions, not even MPEG-1, not even H.261. A "submarine patent" covering these could surface tomorrow.

The most conservative yet practical definition of "patent-free, royalty-free" is no active patent assertions, and VP8 fits that bill.

I try to be practical, you want to argue whether it's possible to guarantee what cannot be guaranteed.