Hello,
 
I have a question to x264. I already posted the question to the x264 developers but i got no answer yet, so I try it here.
 
how is it possible that the diamond search achieves better compression than the Exhaustive Search (both ESA and TESA) for the same video.
Diamond search is faster and achieves a smaller average frame size.
This was tested for different search ranges.
I use fullpixel precision for both algorithms.
The PSNR of Diamond Search is about 34,85 and for TESA  34,96, visually I see no differences.
The test video is a high-motion video which is 355x288 resolution (showing a football match).
 
As ratecontrol method I use CQP and I disabled adaptive quantization.
Furthermore I disable direct motion vector prediction.
As presets I use placebo and zerolatency with the profile high.
 
My first idea was that TESA increases the header information for each frame, but i think this is not the case.
Or is it just a random occurence due to entropy coding (i use CABAC)?
I already read that the exhaustive search doesn't check all possible positions, so maybe the algorithm converges to a local minimum, which is suboptimal for my testvideo.
 
But this is all very unlikely because i tested it for 3 other videos (one high-motion, two low-motion) with the same result.
 
I am lost here and need help cause this is relevant for my master thesis.
In the annotations you find my plots for coding time and frame size.
 
Best regards
 
Felix Schillmaier

Attachment: DiamondSearchCodingTimeFootball0.pdf
Description: Adobe PDF document

Attachment: DiamondSearchFrameSizeFootball0.pdf
Description: Adobe PDF document

Attachment: TESACodingTimeFootball.pdf
Description: Adobe PDF document

Attachment: TESAFrameSizeFootball.pdf
Description: Adobe PDF document

_______________________________________________
Libav-user mailing list
[email protected]
http://ffmpeg.org/mailman/listinfo/libav-user

Reply via email to