Issues

• float_adm: the decouple divides by an RCPSS estimate, so the scores depend on the processor

  Netflix/vmaf#1654 · Filter by author lusoris opened on Oct 1, 2026

• cambi: out-of-bounds read and write on wide, short frames (e.g. 1920x128), C and AVX2 paths

  Netflix/vmaf#1628 · Filter by author lusoris opened on Sep 30, 2026
  1

• speed_temporal: heap-buffer-overflow when speed_prescale > 1 (frame buffers sized with the source height)

  Netflix/vmaf#1626 · Filter by author lusoris opened on Sep 30, 2026

• New HFR models completely broken for AV1

  Netflix/vmaf#1625 · Filter by author stderr-to-devnull opened on Sep 29, 2026
  1

• Correction to my review of #1585: the -15% was measured on a CPU without GDS mitigation

  Netflix/vmaf#1624 · Filter by author lusoris opened on Sep 28, 2026

• Correction to my review of #1586: it changes i4_adm_cm_avx2 and gives 5-7% on Zen 3

  Netflix/vmaf#1623 · Filter by author lusoris opened on Sep 28, 2026

• Correction to my review of #1609: CI was not blocked, and vpblendvb is a 3-9% gain on Zen 3

  Netflix/vmaf#1622 · Filter by author lusoris opened on Sep 28, 2026

• Build failure with llvm-23.1.2 and clang-23.1.2 [error: call to 'swap' is ambiguous]

  Netflix/vmaf#1616 · Filter by author listout opened on Sep 27, 2026
  1

• integer_adm: adm_cm_avx2/avx512 not bit-exact with the C reference for scale-0 coefficients >= 15360; checkasm's ±8000 inputs never reach it

  Netflix/vmaf#1610 · Filter by author lusoris opened on Sep 21, 2026
  1

• adm_cm AVX2/AVX-512 narrow accum_h to float before dividing, unlike accum_v, accum_d and the scalar

  Netflix/vmaf#1608 · Filter by author lusoris opened on Sep 19, 2026

• integer ADM crashes for any frame dimension of 16 or below

  Netflix/vmaf#1607 · Filter by author lusoris opened on Sep 19, 2026
  1

• MSVC Support

  Netflix/vmaf#1597 · Filter by author kylophone opened on Sep 17, 2026
  2

• mirror()-style boundary reflection reads out-of-bounds for width/height < 3 (motion feature, all backends)

  Netflix/vmaf#1580 · Filter by author kjg0724 opened on Aug 13, 2026
  1

• Consider adding a minimal CUDA compile check to CI

  Netflix/vmaf#1576 · Filter by author StormBytePP opened on Aug 11, 2026
  1

• Windows: FFmpeg libvmaf filter creates a mojibake filename for non-ASCII log_path

  Netflix/vmaf#1568 · Filter by author AntaresFeng opened on Jul 20, 2026
  1

• Feature request: CUDA feature extractor for CAMBI

  Netflix/vmaf#1567 · Filter by author teb opened on Jul 18, 2026
  1

• CUDA 16bpc motion kernel applies byte stride as uint16_t element stride — motion2 ~-58%, VMAF ~1.5 points low for all >8-bit input

  Netflix/vmaf#1566 · Filter by author waveletbeam opened on Jul 13, 2026
  1

• integer ADM: CUDA cm diverges from CPU (border mirror + rounding order) and x86 SIMD dwt corrupts the mirrored last output column — fixes attached

  Netflix/vmaf#1564 · Filter by author dsummer opened on Jul 10, 2026
  2

• cuda: motion score diverges from CPU — edge mirror off-by-one in calculate_motion_score_kernel_{8,16}bpc

  Netflix/vmaf#1562 · Filter by author dsummer opened on Jul 10, 2026
  1

• Which (V1) Model ?

  Netflix/vmaf#1561 · Filter by author PhilippeR opened on Jul 10, 2026

• how to use phone model in windows by ffmpeg？

  Netflix/vmaf#1473 · Filter by author yhwu0423 opened on Mar 19, 2026
  1

• Explicitly state stream order in FFmpeg's guide

  Netflix/vmaf#1469 · Filter by author allan-kg opened on Mar 6, 2026
  1

• [question] any ways to use this for audio?

  Netflix/vmaf#1457 · Filter by author RJVB opened on Dec 11, 2025

• VMAF 训练模型中的vmaf_nr_v0.6.1.pkl 能否用其他版本来替换？

  Netflix/vmaf#1454 · Filter by author wujie-test opened on Dec 5, 2025
  1

• ci: build ffmpeg with cuda vmaf

  Netflix/vmaf#1448 · Filter by author kylophone opened on Nov 13, 2025
  1