Unexprected results from compression of bland scene

[cd]

Hey,

I was carrying out an experiment to see how effective HEVC is at compressing video sequences of bland background areas such as white walls.

I captured 1000 uncompressed frames at 1920x1080 and created two seperate YUV files. One of all the raw frames and the other was 1000 copies of the 1st frame.

I then compressed each YUV using the x265 encoder but the results were not what I expected. The final file size for the 1000 different frames was 421 kB however the 1000 repeated frames was 715 kB.

I would have expected these numbers to have been the other way around, can anyone explain what is going on here?

Thanks.

[MeteorRain]

Do you have your command line and encoding log around?

[FranceBB]

The final file size for the 1000 different frames was 421 kB however the 1000 repeated frames was 715 kB.


Did you use reframe? What's your GOP structure?
You can try with something like --ref 16 and a very large --keyint to avoid wasting space on unnecessary Intra frames. Motocompensation will assign Intra to the very first one and then B to the other ones on the 1000 frames sequence.
On the other sequence, however, it won't find as much correlation as all the frames are different.
I don't know how you got the results you got, but your command line would help, as they seem a bit odd.