Skip to content

Tags: rust-av/rav1e

Tags

v0.5.0

Toggle v0.5.0's commit message
CI: Run apt-get update before install

If the apt index in the base image is outdated, fetching packages
may fail with an HTTP 404 error.

p20211026

Toggle p20211026's commit message
Use function pointers instead of `dyn Fn` for `get_func`

p20211019

Toggle p20211019's commit message
CI: Update to libaom to 3.2.0-dmo1

Also update libvmaf to 2.3.0-dmo1.

p20211012

Toggle p20211012's commit message
Add minimal enforcement of padding on input frames

p20211005

Toggle p20211005's commit message
Use SSE for cdef_dist blocks overlapping boundary

Use spatial weights when substituting SSE for cdef_dist.

AWCY results at default preset on subset1:

PSNR Y | PSNR Cb | PSNR Cr | CIEDE2000 |    SSIM | MS-SSIM | PSNR-HVS Y
0.0045 | -0.2649 | -0.2223 |   -0.0184 | -0.0241 | -0.0368 |    -0.0103

PSNR-HVS Cb | PSNR-HVS Cr | PSNR-HVS |    VMAF | VMAF-NEG
    -0.2228 |     -0.2664 |  -0.0142 | -0.0725 |  -0.0506

p20210928

Toggle p20210928's commit message
Use reduced precision in PDF sort

This fixes an underflow bug triggered when the counter is larger than
the previous value. Fixes xiph#2780.

PSNR Y | PSNR Cb | PSNR Cr | CIEDE2000 |   SSIM | MS-SSIM
0.0029 |  0.1181 |  0.0019 |    0.0017 | 0.0059 |  0.0158

PSNR-HVS Y | PSNR-HVS Cb | PSNR-HVS Cr | PSNR-HVS |   VMAF | VMAF-NEG
   -0.0040 |      0.1020 |      0.0126 |  -0.0006 | 0.0959 |   0.0822

p20210921

Toggle p20210921's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
Hide CPU info and encode settings with --quiet (xiph#2810)

v0.5.0-beta.2

Toggle v0.5.0-beta.2's commit message
Bump the v_frame version as additional API got added

p20210914

Toggle p20210914's commit message
Bump the v_frame version as additional API got added

p20210907

Toggle p20210907's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
Simplify `adaptive_scenecut` by not collecting into a Vec for `*_over… (

xiph#2797)

* Simplify `adaptive_scenecut` by not collecting into a Vec for `*_over_tr`

The variables `back_over_tr` and `forward_over_tr` (which have now been renamed to be suffixed with `_count`) were previously initialized by performing an allocation (through the use of `.collect_vec()`) of the filtered values, even though only the *length* of the `Vec` was actually used. We now get the length directly by using `.count()`, which does not perform any allocations, and allows the compiler to make use of SIMD instructions.

Additionally, the previous code essentially did this (simplified for example): `x != 0 && x > 1` (where `x` is a `usize`). This has been simplified to just `x > 1`.

* Further simplification

Do not check if `back_over_tr_count` and `back_deque` are both greater than one, as just checking `back_over_tr_count` is sufficient