Commercial Advance

[Jere_Jones]

<The following information comes from the MythTV camp>

Doing commercial detection on the fly isn't possible when using a hardware encoder. The reason is that the show gets encoded on the card and doesn't go "through" any software filters. If it did then we could do realtime encodes to Divx or other similar things.

MythTV can do real time commercial detection on software encoders only. And since the theory is the same and Sage doesn't support software encoders, we're going to be stuck doing any post processing... um... "post" recording.

But... it's ok! After you use Sage long enough and have enough favorites programed in, you'll rarely watch a program as it is aired. At least I don't.

Jere

[ErsatzTom]

Well, it is true it would need to decode on the fly. That is why it would be expensive in terms of cpu use. Perhaps even totally undoable. But not, I think, technically impossible.

Narflex talks about it in this thread. But he also mentions the difficulty involved in decoding the mpeg on the fly in this thread.

t

[salsbst]

Might it be possible to get the raw video from the 250's Directshow "Preview Pin" at the same time as it is being encoded and output from the encoder pin?

[Narflex]

Actually, I believe there's ways to do it without decoding the MPEG2 stream. There's probably enough information in the MPEG2 encoding that you could use to find scene breaks without decoding the stream. Its much more complicated because you're working in the encoded domain instead of the raw domain.

You'd probably have to do some kind of partial decode, but you could almost definitely avoid doing the IDCT back into the spatial domain and do your work in the frequency domain because energy is conserved across time/freq domains and black frames are essentially low energy frames. (my specialization was DSP back in college) You could also limit your decoding to only a few macroblocks if you're detecting high energy to avoid having to decode an entire frame.

Keep in mind that's all just theory.....I've never done an implementation like that.

[malore]

Originally posted by ErsatzTom
Also, I don't suppose the output format is close to, um, other software output so as to make it work with VideoReDo?

Not directly, but it wouldn't take much work to convert from the simple frame numbers my program outputs to the microsecond time codes that VideoReDo can use.

[ErsatzTom]

Originally posted by malore
Not directly, but it wouldn't take much work to convert from the simple frame numbers my program outputs to the microsecond time codes that VideoReDo can use.

Or, once we get this going maybe the folks who make VideoReDo would be willing to support our format as well. It couldn't be too much work on their part and they would get (even more) good press on our forums.

t

[ErsatzTom]

Originally posted by Narflex
Actually, I believe there's ways to do it without decoding the MPEG2 stream. There's probably enough information in the MPEG2 encoding that you could use to find scene breaks without decoding the stream. Its much more complicated because you're working in the encoded domain instead of the raw domain.

Don't suppose you already know where to find additional resources for this type of undertaking? Thanks!

t

[ErsatzTom]

Oh, and because I didn't say it before...

THANK YOU, THANK YOU, THANK YOU malore!!!

t

[Fluffdaddy]

Originally posted by ErsatzTom
Oh, and because I didn't say it before...

THANK YOU, THANK YOU, THANK YOU malore!!!

t

Dttio

[malore]

I anyone missed it, I release a beta version for people to try.