[FFmpeg-user] I found the bugs

[Mark Filipak]

On 17/06/2024 18.05, Paul B Mahol wrote:
> On Tue, Jun 18, 2024 at 12:00 AM Mark Filipak <markfilipak.imdb at gmail.com>
> wrote:
> 
>> On 17/06/2024 17.18, Paul B Mahol wrote:
>>> On Mon, Jun 17, 2024 at 8:29 PM Mark Filipak <markfilipak.imdb at gmail.com
>>>
>>> wrote:
>>>
>>>> On 17/06/2024 13.50, Paul B Mahol wrote:
>>>>> On Mon, Jun 17, 2024 at 5:35 PM Mark Filipak <
>> markfilipak.imdb at gmail.com
>>>>>
>>>>> wrote:
>>>>>
>>>>>> On 17/06/2024 03.16, Paul B Mahol wrote:
>>>>>>> On Mon, Jun 17, 2024 at 3:16 AM Mark Filipak <
>>>> markfilipak.imdb at gmail.com
>>>>>>>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> To expand on my idea:
>>>>>>>>
>>>>>>>> The video is this:
>>>>>>>> I P B B P B B P B B I P B B..
>>>>>>>> I want to change it to this:
>>>>>>>> I B B P B B P B B P I B B P..
>>>>>>>>
>>>>>>>
>>>>>>> If this is what I think it is, your are wasting your time.
>>>>>>
>>>>>> It is possible. But can FFmpeg do it? You see, Paul, I read in several
>>>>>> places from seemingly
>>>>>> authoritative people that professional videos are PTS-ordered. I've
>>>> since
>>>>>> discovered that
>>>>>> professional videos are almost always DTS-ordered -- I now have a
>> packet
>>>>>> browser and have seen it to
>>>>>> be true. However, I have seen professional videos that are PTS-ordered
>>>> but
>>>>>> I didn't think too much
>>>>>> about it at the time.
>>>>>
>>>>> PTS order is order once frames are decoded and going to be presented to
>>>>> user, thats why it is called presentation timestamps.
>>>>
>>>> Yes of course.
>>>>
>>>>>> My idea is to take the 4-second video -- DTS-ordered -- that I sent to
>>>>>> trac, change it to
>>>>>> PTS-ordered, and see if '-vf showinfo' and '-f show_frames' respond
>>>>>> correctly. It's a long shot, but
>>>>>> it might show something useful -- what if PTS-order made showinfo
>>>>>> show_frames and framecrc all
>>>>>> agree? Wouldn't that be something interesting?
>>>>>
>>>>> If decoder follows specifications for decoding bitstreams correctly
>> than
>>>>> changing order of raw encoded frames as stored in bitstream will almost
>>>>> every-time cause broken decoding.
>>>>>
>>>>>>> Its nonsense to change order of encoded frames, no manipulation of
>>>> clock
>>>>>>> can fix that.
>>>>>>
>>>>>> No, no. Manipulation of PCR. Fix up the PCRs so that the result is
>> truly
>>>>>> PTS-ordered.
>>>>>
>>>>> If this is at format container layer, I dunno how would that work even
>> if
>>>>> its is possible to be useful.
>>>>
>>>> The container would contain a PES packet stream with PCRs like any other
>>>> PES packet stream. It would
>>>> work like any PTS-ordered stream does. You know, there _are_ PTS-ordered
>>>> streams. And you know, MPEG
>>>> is silent on the matter of what physical order frames are in the stream,
>>>> only that they must be
>>>> given to the decoder in DTS-order.
>>>>
>>>> Look, I agree that it would be nonsense. I want to turn the DTS-ordered
>>>> stream into a PTS-ordered
>>>> stream merely as a test: '-f show_frames' appears to be getting DTSes
>> and
>>>> PTSes mixed up, then
>>>> applying a fix that results in completely new DTSes. If it treats the
>>>> PTS-ordered stream
>>>> differently, that will say a lot. What is says depends on what it does.
>>>> Capiche?
>>>>
>>>> I thought I could do it by demuxing, then moving them like you can move
>>>> frames in a filter_complex,
>>>> and then remuxing, but I don't know how to do that to PES packets
>> (instead
>>>> of decoded pictures) and
>>>> I don't know how to rewrite PCRs.
>>>>
>>>
>>> But aren't you working with HEVC and not MPEG video?
>>
>> I'm working with AVC-M2TS, then transcoding-remuxing to HEVC-MP4.
>
> So you operate with bsf filters to create HEVC output?

No. Let me explain.

My original project is to edit and join the 4 episodes of Bergman's FANNY OCH ALEXANDER (5h11m). 
That's going to be HEVC-MP4 -- actually I've already done it but it has gaps at the splices. For 
that Bergman project I transcode of course.

I had very carefully planned out the 7 cuts and 3 IDR splices using '-bsf noise' and '-bsf setts'. I 
have the scripts (Windows) if you care to see them -- if I may say so myself, they're quite elegant.

Of course, I tried to use '-ss' and '-to' at first, but they made a mess.

By the way, '-i "concat:...|..."' is not the problem many people think it is. The problem is 
actually the timestamp bug I found.

Okay, back to the subject: I found the timestamp bug and clipped out a 4 second sample that spans 
the first splice. That's what I gave to trac. It is AVC-M2TS, the same as the Criterion original. 
It's for test cases of that 4 second sample that I want to create a PTS-order version.

I hope that's clear, now. I apologize for the confusion.