Using target mode, probes are getting much lower VMAF scores than the final encode. I am using --probe-video-params copy, I expected they'd be closer.
This is my command
av1an -i intermediate.mkv \
-o av1an_target93.mkv \
--temp target93_temp \
-l target93_log \
--encoder svt-av1 \
--target-quality 93 \
--chunk-method lsmash \
--concat mkvmerge \
--probe-video-params copy \
--vmaf \
-v "--preset 2 --keyint 420 --enable-qm 1 --enable-variance-boost 1" \
--pix-format yuv420p10le \
-x 0
Which gives this representative output.
2026-02-10T01:30:02.542162Z DEBUG encode_chunk{worker_id=0 total_chunks=315 chunk_index="00192"}: av1an_core::target_quality: chunk 00192: Target=92.07-93.93, Metric=vmaf, P-Rate=1, 352 frames, P-Video-Params: --preset 2 --keyint 420 --enable-qm 1 --enable-variance-boost 1
TQ-Probes: [(15.00, 88.98), (23.75, 85.31), (32.50, 80.08)]
Final Q=15.00, Final Score=88.98
2026-02-10T01:31:21.727749Z DEBUG encode_chunk{worker_id=1 total_chunks=315 chunk_index="00147"}: av1an_core::target_quality: chunk 00147: Target=92.07-93.93, Metric=vmaf, P-Rate=1, 352 frames, P-Video-Params: --preset 2 --keyint 420 --enable-qm 1 --enable-variance-boost 1
TQ-Probes: [(15.00, 90.07), (23.75, 85.77), (32.50, 79.45)]
Final Q=15.00, Final Score=90.07
2026-02-10T01:32:18.551723Z DEBUG encode_chunk{worker_id=0 total_chunks=315 chunk_index="00133"}: av1an_core::target_quality: chunk 00133: Target=92.07-93.93, Metric=vmaf, P-Rate=1, 348 frames, P-Video-Params: --preset 2 --keyint 420 --enable-qm 1 --enable-variance-boost 1
TQ-Probes: [(16.25, 92.24), (17.50, 91.53), (23.75, 87.59), (32.50, 79.93)] Early Skip Within Tolerance
Final Q=16.25, Final Score=92.24
It's having to go into the teens to get close to my target VMAF of 93.
On average, the Q=16.8 for a VMAF of 91.5
When I run vmaf on the final file, I get 96.2
When I encode to CRF 30 using the same parameters, I get a VMAF of 93.8
My source is an already bob deinterlaced PAL DVD at 50fps.
Stream #0:0: Video: ffv1, yuv420p10le(tv, progressive), 720x576, 50 fps, 50 tbr, 1k tbn
Is this expected, or am I doing something wrong?
Using target mode, probes are getting much lower VMAF scores than the final encode. I am using --probe-video-params copy, I expected they'd be closer.
This is my command
Which gives this representative output.
It's having to go into the teens to get close to my target VMAF of 93.
On average, the Q=16.8 for a VMAF of 91.5
When I run vmaf on the final file, I get 96.2
When I encode to CRF 30 using the same parameters, I get a VMAF of 93.8
My source is an already bob deinterlaced PAL DVD at 50fps.
Stream #0:0: Video: ffv1, yuv420p10le(tv, progressive), 720x576, 50 fps, 50 tbr, 1k tbn
Is this expected, or am I doing something wrong?