Quality decrease in 1.2.0 comparing to 1.1.0

[GoogleCodeExporter]

>What is the expected behavior? What do you see instead?
We expect that quality in same or better than 1.1.0, but quality is worse.

>What version are you using? On what operating system?
libvpx v1.2.0, Windows and other OS

currently we are using libvpx v1.1.0 , after VP9 announcement we decided to try 
1.20
News about 1.2.0 contained "In general, quality is consistent with the Eider 
release".
But after quality comparison, we found that Y-PSNR decreased on several and 
performance is approximately same.

Coder settings:

    cfg.g_usage = 0;
    cfg.g_profile = 0;
    cfg.g_w = 640;
    cfg.g_h = 480;
    cfg.g_timebase.num = 1;
    cfg.g_timebase.den = 30;
    cfg.rc_target_bitrate = 50...1000;
    cfg.rc_end_usage = VPX_CBR;
    cfg.rc_min_quantizer = 4;
    cfg.rc_max_quantizer = 63;
    cfg.rc_buf_initial_sz = 2000;
    cfg.rc_buf_optimal_sz = 2000;
    cfg.rc_buf_sz = 3000;
    cfg.rc_dropframe_thresh = 0;
    cfg.rc_overshoot_pct = 0;
    cfg.rc_undershoot_pct = 0;
    cfg.g_pass = VPX_RC_ONE_PASS;
    cfg.g_error_resilient = 0;
    cfg.g_threads = 1;
    cfg.kf_mode = VPX_KF_AUTO;
    cfg.kf_max_dist = 1000;
    cfg.g_lag_in_frames = 0;
    cfg.ts_number_layers = 1;

    vpx_codec_control_(pCodec, VP8E_SET_CPUUSED, 5);
    vpx_codec_control_(pCodec, VP8E_SET_STATIC_THRESHOLD, 800);

    Other parameters are default from vpx_codec_enc_config_default.

    encoding
    vpx_codec_encode(ctx, img, frame_cnt, 1, 0, VPX_DL_GOOD_QUALITY);

We will try to attach sample video stream that reproduces problems - 640x480@30 
(it is used, for example,  in  MSU CS  GRAPHICS & MEDIA LAB VIDEO GROUP 
comparisons) 

Same settings do not cause problems on foreman (CIF@30)  sequence

additional info
1) We compared basic version of 1.1.0 and 1.2.0 without additional modifications
2) Same this reproduces woth basic 1.1.0 and current 1.2.0 snapshot from GIT  
(01/09/2013)

Original issue reported on code.google.com by webmas...@trueconf.com on 4 Sep 2013 at 12:32

[GoogleCodeExporter]

in cfg.rc_target_bitrate = 50...1000 range decrease in Y-PSNR is about 0.3dB

here is the sample sequence
https://drive.google.com/folderview?id=0B9Alqmk4UAdIRU5Fcm9lcmlRNk0&usp=sharing

Original comment by webmas...@trueconf.com on 4 Sep 2013 at 1:10

[GoogleCodeExporter]

I have the same problem on some video streams!

Original comment by math.bes...@gmail.com on 25 Sep 2013 at 3:30

[GoogleCodeExporter]

You are lucky since rt only version does not ever compile

Original comment by dron...@gmail.com on 25 Sep 2013 at 3:40

[GoogleCodeExporter]

Can you please provide the command line you are using completely?

Original comment by jimbankoski@google.com on 17 Oct 2013 at 10:35

• Changed state: Assigned

[GoogleCodeExporter]

currently we are using libvpx in C++ code

Original comment by webmas...@trueconf.com on 23 Oct 2013 at 3:56

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

--codec=vp8
--width=640
--height=480
--fps=30000/1000
--target-bitrate=50...1000
--good
--cpu-used=5
--end-usage=cbr
--buf-initial-sz=2000
--buf-optimal-sz=2000
--buf-sz=3000
--undershoot-pct=0
--overshoot-pct=0
--passes=1
--kf-max-dist=1000
--auto-alt-ref=0
--lag-in-frames=0
--threads=1
--drop-frame=0
--resize-allowed=0
--error-resilient=0
--static-thresh=800
--profile=0
--min-q=4
--max-q=63 

Original comment by webmas...@trueconf.com on 24 Oct 2013 at 3:46

[GoogleCodeExporter]

We found a git patch when this bug was introduced: 21 August 2012
patch name: Add biasing to ZEROMV for videos with static background 
by Yunqing Wang
While slighly increasing speed, this patch lowers quality visually and 
noticeable in measurements.

Original comment by webmas...@trueconf.com on 5 May 2014 at 2:18

[GoogleCodeExporter]

Is this still relevant? or has it been fixed already?

Original comment by ya...@google.com on 10 Jul 2015 at 9:28

[GoogleCodeExporter]

It's up to you, since git revert works flowessly in our cases

Original comment by dron...@gmail.com on 10 Jul 2015 at 9:31

[GoogleCodeExporter]

The patch was added to visually help video conferencing encoding in some cases. 
After this initial patch, several related patches were put in to fix/improve 
the method. You may want to check out the newest code to see if the issue still 
exists.

Original comment by yunqingw...@google.com on 10 Jul 2015 at 10:27