Detectads operation.

[Trev]

I am running detectads in chaseget mode. Having built up some confidence of correct cropping, I have also set the crop after detection on.
When I am replaying the video, I get a bit of picture break up at the crop points. Is this normal? It's not a major thing, just asking.

 

[Black Hole]

It the crop did not occur at an I Frame, it would.

 

[Trev]

Ok, so that's the techie explanation but begs the question, "Why isn't it cropping at the I Frame?"
Should it?

 

[Black Hole]

Don't know. Read the nicesplice topic.

I believe all the detectads process does is set bookmarks - they have a very limited time resolution. Whether the nicesplice process looks for the nearest I frame to a bookmark, or just crops at the bookmark, I can't recall.

For those who don't know: video compression algorithms produce a stream of P frames interspersed with I frames. The I frame data decompresses to a complete still image, and the P frame data only describes the differences in subsequent frames. I frames occur at a preset maximum interval and any time the differences accumulate to make it worth putting one in (a change of scene for example, when the differences from the previous frame will be huge). Simply cropping a data stream between I frames means there is no baseline for the P frames until an I frame comes along. Editing video to frame resolution requires decompressing the data to fully express the detail of each frame individually, and is a compute-intensive process unreasonable to attempt on the HDR-FOX.

 

[MymsMan]

NiceSplice does look for an I-frame within a couple of seconds of the crop point.
There can also be lack of sound for a couple of seconds after an ad break and problems with blocky fast forwarding across a cropped ad break.
These are due to the way nicesplice works rather than detectads and I need to find time to try and understand the code to see if it can be improved.
The switch between programme and ads should always be a significant scene change so should require an I-frames so why it sometimes causes problems needs further investigation.

 

[Black Hole]

The switch between programme and ads should always be a significant scene change

That's as may be, but if the only way to communicate cues between detectads and nicesplice is by bookmarks, the resolution is lost.

 

[MymsMan]

That's as may be, but if the only way to communicate cues between detectads and nicesplice is by bookmarks, the resolution is lost.

But there should be an I-frame within a second for nicesplice to find restoring the resolution

 

[Trev]

I assumed that it was something to do with the actual cropping, and now understand a little bit more about the difficulties in making a perfect crop.
As I said in my OP, it's not a biggie, in fact not really a problem at all. I just wondered what was causing it.
However, if Myms, the perfectionist that he is, feels like it ..........