All pull requests

Pull requests

• libvmaf: fix y4m direct read with mixed 8/10-bit inputs

  #1682·kslu-aom opened Oct 12, 2026·

  ·
  1

• psnr: report the per-frame cap for an APSNR plane without error

  #1666·lusoris opened Oct 2, 2026·

  ·

• feature/ms_ssim: take the magnitude of the structure term before pow()

  #1665·lusoris opened Oct 2, 2026·

  ·
  1

• python: keep motion_force_zero a bool across models in call_vmafexec()

  #1663·lusoris opened Oct 2, 2026·

  ·

• cuda: add an approximate float32 CIEDE2000 extractor

  #1661·BardieJoensen opened Oct 2, 2026Contributor·

  ·

• cuda: add PSNR, MSE and APSNR extraction

  #1660·BardieJoensen opened Oct 2, 2026Contributor·

  ·
  1

• tools/vmaf: release everything on an error exit of main()

  #1659·lusoris opened Oct 2, 2026·

  ·

• feature/cambi: store the integer options in int fields

  #1658·lusoris opened Oct 2, 2026·

  ·

• feature/adm: divide in the float decouple instead of estimating the reciprocal

  #1655·lusoris opened Oct 1, 2026·

  ·
  1

• libvmaf: release both pictures on every return of vmaf_read_pictures()

  #1652·lusoris opened Oct 1, 2026·

  ·
  1

• feature/adm/cuda: take the scale 0 denominator shift from the host

  #1651·lusoris opened Oct 1, 2026·

  ·

• feature/adm/cuda: decide the one degree angle in the scale 0 decouple as the CPU does

  #1650·lusoris opened Oct 1, 2026·

  ·

• feature/adm/cuda: clamp the scale 0 contrast masking neighbours at the border

  #1649·lusoris opened Oct 1, 2026·

  ·

• feature/adm/cuda: round the scale 0 and scale 1 to 3 contrast masking sums once per row

  #1648·lusoris opened Oct 1, 2026·

  ·

• feature/adm/cuda: read the contrast masking neighbours of scales 1 to 3 as the CPU does

  #1647·lusoris opened Oct 1, 2026·

  ·

• cuda: return an error from vmaf_cuda_buffer_alloc() when the device is out of memory

  #1646·lusoris opened Oct 1, 2026·

  ·

• cuda: unload the kernel modules and destroy the streams of the motion, vif and adm extractors

  #1645·lusoris opened Oct 1, 2026·

  ·

• cuda: use the CPU's reflect-101 edge mirror in the motion kernels

  #1644·lusoris opened Oct 1, 2026·

  ·

• test/predict: compare against a double in test_post_process_feature_from_another

  #1643·lusoris opened Oct 1, 2026·

  ·

• feature/adm: reject frames below 17 pixels in either dimension at init

  #1642·lusoris opened Oct 1, 2026·

  ·
  1

• libvmaf: open the output file by its UTF-8 name on Windows

  #1640·lusoris opened Oct 1, 2026·

  ·

• libvmaf: reject picture indices that do not increase in vmaf_read_pictures()

  #1639·lusoris opened Oct 1, 2026·

  ·
  2

• tools: keep a drive-letter colon inside the --model path

  #1638·lusoris opened Oct 1, 2026·

  ·
  1

• feature/adm: keep get_best15_from32() inside its defined range on AVX2

  #1635·lusoris opened Oct 1, 2026·

  ·
  1

• feature/adm: truncate the gain-limited sample in the AVX2 and AVX-512 decouple kernels

  #1633·lusoris opened Oct 1, 2026·

  ·
  1