Google VP9 "Next Generation Open Video" information posted

[LigH]

New uploads (MSYS2; MinGW32 / MinGW64: GCC 13.2.0):

VPx v1.14.0-162-gd4959f982

SVT-VP9 0.3.0-3ecdf8f

[LigH]

New uploads (MSYS2; MinGW32 / MinGW64: GCC 14.1.0):

VPx v1.14.1-318-g3219f76c

[LigH]

New uploads (MSYS2; MinGW32 / MinGW64: GCC 14.2.0):

VPx v1.14.1-359-gc6de95ce

[takla]

For anyone still interested in VP9:
I wanted to point out, that it has, in recent times, recieved multiple new AQ-Modes, which were not mentioned in any patchnotes, iirc.

-aq-mode X

1 = VARIANCE_AQ
2 = COMPLEXITY_AQ
5 = PERCEPTUAL_AQ
6 = PSNR_AQ
7 = LOOKAHEAD_AQ  <-- requires -auto-alt-ref 1 (or greater) which itself requires 2-pass

Source

[LigH]

New uploads (MSYS2; MinGW32 / MinGW64: GCC 14.2.0):

VPx v1.15.0-22-g66d339b3

SVT-VP9 0.3.0-1feb760

[Blue_MiSfit]

Does anyone have recommendations for quick and dirty libvpx-vp9 encoding parameter settings for relatively fast file based encoding for a ~4 Mbps capped CRF 1080p SDR editorial proxy use case?

A certain popular NLE doesn't support HEVC or AV1 for proxies, but does support AVC and VP9, so I've been asked to look into how libvpx-vp9 compares with x264 for this use case.

Are there any settings that aren't defaults but probably should be? Is there current guidance on lag-in-frames, auto-alt-ref, and row-mt? I'm paraphrasing as I forget the actual parameter names

[GeoffreyA]

ffmpeg -i INPUT -c:v libvpx-vp9 -quality good -cpu-used 3 -crf 18 -b:v 4000k OUTPUT

[takla]

Release v1.15.0 Wigeon Duck

2024-10-31 v1.15.0 "Wigeon Duck"
This release includes new codec control for key frame filtering, more Neon
optimizations, improvements to RTC encoding and bug fixes.

- Upgrading:
This release is ABI compatible with the previous release.

Temporal filtering improvement that can be turned on with the new codec
control VP9E_SET_KEY_FRAME_FILTERING, which gives 1+% BD-rate saving with
minimal encoder time increase.

libwebm is upgraded to libwebm-1.0.0.31-10-g3b63004

- Enhancement:
Neon optimization speed up
1-3% speed up across speed 5 to 10 for RTC
3% speed up for speed 0 and 1 for VoD in standard bitdepth
3% and 7% speed up for speed 0 and 1 respectively for VoD in high bitdepth
Scene detection is allowed for all RTC speeds (>=5)
Support profile guided optimizations

Delta quantization parameters for UV channels for vp8 is supported in RTC
rate control library

Rate control parameters are reset and maximum QP is enforced on scene
changes in SVC when there is no inter-layer prediction

- Bug fixes:
Fix to Uninitialized scalar variable in `vp9_rd_pick_inter_mode_sb()`
Fix to Integer-overflow in `resize_multistep`
Fix to Heap-buffer-overflow in `vpx_sad64x64_avx2`
Fix to Crash in `vpx_sad8x8_sse2`
Fix to Assertion in `write_modes`
Support profile guided optimizations
Fix to Integer-overflow in `encode_frame_to_data_rate`
Fix to Integer-overflow in `vp9_svc_check_reset_layer_rc_flag`
Fix to core dump error from /usr/bin/tools/tiny_ssim --help
Fix to use-of-uninitialized-value in `vp9_setup_tpl_stats`
Fix to Undefined-shift in `vp9_cyclic_refresh_setup`
Fix to redundant `&& __GNUC__` preproc check
Fix to valgrind warning in EncodeAPI.OssFuzz69906
Fix to Index-out-of-bounds in `vp8_rd_pick_inter_mode`
Fix to Integer-overflow in `vp8_pick_frame_size`
Fix to Use-of-uninitialized-value in `vpx_codec_peek_stream_info`
Fix to log clutters with the message "Warning: Desired height too large"
Fix to Integer-overflow in `vp9_svc_adjust_avg_frame_qindex`

Fix to integer overflows caused by huge target bitrate, frame rate, or
g_timebase numerator or denominator

Fix to missing license headers
Fix to build failure for Android Armv7
Fix to integer overflows in image helpers
Fix to Integer-overflow in `vp9_calc_iframe_target_size_one_pass_cbr`
Fix to Heap-buffer-overflow in `vp9_pick_inter_mode`
Fix to Segv in `vp9_multi_thread_tile_init`
Fix to Use-of-uninitialized-value in `vp9_row_mt_sync_mem_dealloc`
Fix to Crash in `mbloop_filter_vertical_edge_c`
Fix to Check failed in CheckUnwind
Fix to Heap-buffer-overflow in `write_modes_b` and `vpx_write`
Fix to Possible signed integer overflow found in `vpx_codec_encode`
Fix to build conflicts between Abseil and libaom/libvpx in Win ARM64 builds
Fix to build failures on aarch64
Fix to Data race in libvpx ARM NEON
Fix to Heap-buffer-overflow in `scale_plane_1_to_2_phase_0`
Fix to integer overflow in `encode_mb_row`
Fix to Floating-point-exception in `vp8_pick_frame_size`
Fix to Heap-buffer-overflow in `vp9_enc_setup_mi`
Fix to build failure with --target=arm64-win64-vs17
Fix to heap-buffer-overflow write in `vpx_img_read()`
Fix to C vs armv8-linux-gcc encode mismatches for `y4m_360p_10bit_input`
Fix to Null-dereference READ in `ml_predict_var_rd_partitioning`
Fix to Heap-buffer-overflow in `vpx_scaled_2d_ssse3`
Fix to Crash in `convolve_horiz`
Fix to Ill in `vpx_scaled_2d_ssse3`
Fix to Global-buffer-overflow in `cost_coeffs`

Source

[takla]

Originally Posted by Blue_MiSfit

Are there any settings that aren't defaults but probably should be? Is there current guidance on lag-in-frames, auto-alt-ref, and row-mt? I'm paraphrasing as I forget the actual parameter names

-lag-in-frame is only used in 2pass, and defaults to the maximum of 25, so no reason the specify it

-auto-alt-ref also only works in 2pass. Personally, I'd never use it. It blurs frames too much.

-row-mt 1 can speed things up

And don't forget to set a keyint value. -g 240 for example.

[LigH]

New upload (MSYS2; MinGW32 / MinGW64: GCC 14.2.0):

VPx v1.15.0-28-g6f0c446c

[takla]

Originally Posted by oibaf

2025-02-10 v3.12.0

Thats for libaom-av1

So, wrong thread.
Should have posted it here.

[LigH]

New upload (MSYS2; MinGW32 / MinGW64: GCC 14.2.0):

VPx v1.15.0-69-g027bbee3