--tune zerolatency does not work with --crf

[Former user]

When I use both of these options, the bitrate stays very high regardless of specified crf.

[Aarthi Thirumalai]

Can you provide us more information on your source, x265 version, cmd line (output log -csv file) is possible? When tested with standard HD, 4k videos, crf with zero latency works as expected.

[Former user]

I am encoding a short segment from parkjoy at 720p resolution, available from derf's test media collection: https://media.xiph.org/video/derf/

crf 10:

Command, Date/Time, Elapsed Time, FPS, Bitrate, Y PSNR, U PSNR, V PSNR, Global PSNR, SSIM, SSIM (dB), I count, I ave-QP, I kpbs, I-PSNR Y, I-PSNR U, I-PSNR V, I-SSIM (dB), P count, P ave-QP, P kpbs, P-PSNR Y, P-PSNR U, P-PSNR V, P-SSIM (dB), B count, B ave-QP, B kpbs, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), Version
 -r park_joy_short.y4m-10.y4m --preset slow --tune zerolatency --threads 1 --min-keyint 256 --keyint 256 --no-scenecut --crf=10 --csv park_joy_short.y4m-10.csv -o park_joy_short.y4m-10.x265 /home/ec2-user/sets/video-1-short/park_joy_short.y4m, Tue Oct 21 16:00:17 2014, 270.60, 0.18, 320731.16, -, -, -, -, -, -,1     , 6.74, 305536.80, -, -, -, -, 49    , 0.34, 321041.21, -, -, -, -, -, -, -, -, -, -, -,  1.2+248-a2fd8a71de61

crf 45:

Command, Date/Time, Elapsed Time, FPS, Bitrate, Y PSNR, U PSNR, V PSNR, Global PSNR, SSIM, SSIM (dB), I count, I ave-QP, I kpbs, I-PSNR Y, I-PSNR U, I-PSNR V, I-SSIM (dB), P count, P ave-QP, P kpbs, P-PSNR Y, P-PSNR U, P-PSNR V, P-SSIM (dB), B count, B ave-QP, B kpbs, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), Version
 -r park_joy_short.y4m-45.y4m --preset slow --tune zerolatency --threads 1 --min-keyint 256 --keyint 256 --no-scenecut --crf=45 --csv park_joy_short.y4m-45.csv -o park_joy_short.y4m-45.x265 /home/ec2-user/sets/video-1-short/park_joy_short.y4m, Tue Oct 21 16:00:12 2014, 266.59, 0.19, 316925.69, -, -, -, -, -, -,1     , 41.89, 7207.60 , -, -, -, -, 49    , 0.34, 323246.44, -, -, -, -, -, -, -, -, -, -, -,  1.2+248-a2fd8a71de61

These videos are very short (2 seconds). Using just --bframes=0 instead of --tune exhibits the expected behavior, but is not causal.

[Aarthi Thirumalai]

The version (1.2+248-a2fd8a71de61) you are using is pretty old. tune=zerolatency was broken in that. the issues have been fixed in eba786bbe4b5. Pls update to the commit eba786bbe4b5 or later to that, to test for zero latency feature.

I ran a test on the current tip : crf 45: Command, Date/Time, Elapsed Time, FPS, Bitrate, Y PSNR, U PSNR, V PSNR, Global PSNR, SSIM, SSIM (dB), I count, I ave-QP, I kpbs, I-PSNR Y, I-PSNR U, I-PSNR V, I-SSIM (dB), P count, P ave-QP, P kpbs, P-PSNR Y, P-PSNR U, P-PSNR V, P-SSIM (dB), B count, B ave-QP, B kpbs, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), Version -o out.x265 --crf 45 --tune zero-latency --log 3 --threads 1 --no-scenecut --csv parkjoy_45.csv C:\x265\testFiles\park_joy_1080p50.y4m, 10/26/14 23:28:21, 37.33, 13.40, 231.44, -, -, -, -, -, -,2 , 45.52, 8994.60 , -, -, -, -, 498 , 51.03, 196.25 , -, -, -, -, -, -, -, -, -, -, -, 1.3+618-c15bb6a0d01f

crf 10: Command, Date/Time, Elapsed Time, FPS, Bitrate, Y PSNR, U PSNR, V PSNR, Global PSNR, SSIM, SSIM (dB), I count, I ave-QP, I kpbs, I-PSNR Y, I-PSNR U, I-PSNR V, I-SSIM (dB), P count, P ave-QP, P kpbs, P-PSNR Y, P-PSNR U, P-PSNR V, P-SSIM (dB), B count, B ave-QP, B kpbs, B-PSNR Y, B-PSNR U, B-PSNR V, B-SSIM (dB), Version -o out.x265 --crf 10 --tune zero-latency --log 3 --threads 1 --no-scenecut --csv parkjoy_10.csv C:\x265\testFiles\park_joy_1080p50.y4m, 10/26/14 23:52:08, 311.56, 1.60, 204085.65, -, -, -, -, -, -,2 , 10.60, 499535.00, -, -, -, -, 498 , 18.32, 202899.10, -, -, -, -, -, -, -, -, -, -, -, 1.3+618-c15bb6a0d01f

[Steve Borho]

marking invalid, since this was long since fixed