Why does x264 scene change detection cluster I-frames?

[Avenger007]

Scene change detection (scd) clusters I-frames during white transitions in this (http://www.filefactory.com/file/e0d5ff/) clip from the Zerg Reveal Trailer.
I tried using scd sensitivity of 30 but there were still 2 or 3 I-frames clustered together and I-frames were removed from some places.
BTW I use Avinaptic to view frame types.

Shouldn't there be some weighted value attached to the scd sensitivity (when within --min-keyint) to avoid I-frame clustering?

Settings:
--pass 2 --bitrate 2000 --keyint 300 --min-keyint 30 --ref 16 --mixed-refs --no-fast-pskip --bframes 8 --b-pyramid --b-rdo --bime --weightb --direct auto --subme 7 --trellis 2 --partitions all --8x8dct --me umh --merange 32 --threads auto --thread-input --aq-strength 1.25

[Gabriel_Bouvigne]

Might be I frames, but that doesn't mean that they are key frames. I frame doesn't imply recovery point/IDR.

[Avenger007]

Might be I frames, but that doesn't mean that they are key frames. I frame doesn't imply recovery point/IDR.
Thanks for pointing that out. :)

I guess I should change the thread title from "Why does x264 scene change detection override --min-keyint?" to "Why does x264 scene change detection cluster I-frames?".

[Selur]

afaik --min-keyint should not limit I-frame, but IDR-Frames, so has nothing to do with 'I-Frame clusters'

[Avenger007]

afaik --min-keyint should not limit I-frame, but IDR-Frames, so has nothing to do with 'I-Frame clusters'
OK, I guess I should cross that part out.

[GodofaGap]

Is there a reason why you want to avoid it? I-frames are not necessarily less efficient than P or B.

[Avenger007]

I thought it would have used some sort of weighted prediction, it's just that the transition brightens instead of darkens the image.
There is also another transition that darkens and that was handled properly (1 I-frame).

[Dark Shikari]

I thought it would have used some sort of weighted prediction, it's just that the transition brightens instead of darkens the image.
There is also another transition that darkens and that was handled properly (1 I-frame).Good weighted prediction is a hard problem; I've only seen one encoder handle it properly ever.

[Dust Signs]

@Dark Shikari: which one?

Dust Signs

[Dark Shikari]

@Dark Shikari: which one?Ateme.

[Dust Signs]

You mean in their Kyrion File Encoder?

Dust Signs

[Dark Shikari]

You mean in their Kyrion File Encoder?

Dust SignsI have only seen output from their hardware encoders in the form of broadcast television streams, but AFAIK their hardware and software encoders use similar algorithms.