GitLab

When run with --enable-ec_adapt, forward updates to the tx_size_prob
 array will not be made.

objective-1-fast:

master-ec_adapt-e568e6f0@2017-02-17T15:30:27.109Z ->
 tx_size-ec_adapt@2017-02-17T15:31:49.504Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0113 | -0.0112 | -0.0113 |  -0.0114 | -0.0114 | -0.0114 |    -0.0112

master-ec_adapt-robust-e568e6f0@2017-02-17T20:29:45.315Z ->
 tx_size-ec_adapt-robust@2017-02-17T20:33:06.405Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0244 | -0.0243 | -0.0243 |  -0.0245 | -0.0245 | -0.0245 |    -0.0243

Change-Id: I5873c5d14e3f3868997775f85fa743d2d55d81b7

Replace obselete size "4" with correct one "tile_size_bytes". This
fixes a decoder crash.

BUG=aomedia:352

Change-Id: I5b1c606a4d5be946acc68b3cfe0d0603a5591c44

Adds functinoality for least-squares, RANSAC as well as encoding and
decoding with new constrained homographies that warp blocks to horizontal
and/or vertical trapezoids. This is for future experimentation. None
of the models are actually enabled in the code.

Change-Id: I1936018c6b11587d6fd83c3a2c63548cb641b33f

Allows DomainTxfm filters to be turned off for experimentation.
Also expands the parameter set for the Self guided filters.

Change-Id: I68fdb8e079a2464d80b3a4a990005c49baaaf0b8

This fixes a decoding failure issue of adaptive scan order in the
error resilient mode.

BUG=aomedia:353

Change-Id: I4b09f168cd0e0efc843a05f82d7e1c05173ee629

This commit makes the adaptive scan order system support multi-
thread encoding. It fixes unit test failure issue associated with
AV1/AVxEncoderThreadTest.EncoderResultTest/0.

BUG=aomedia:353

Change-Id: I61cbf9531c8deab97fb3bb17428d0b2a63cf309a

Noticed that some ALTREF_FRAMEs could have used compound modes for its
prediction but have been labeled as SINGLE_REFERENCE mode in the frame
header. This experiment is to remove the COMPOUND_REFERENCE mode from
the frame-level reference mode choices and only leave SINGLE_REFERENCE
and REFERENCE_MODE_SELECT the two choices in the frame header.

When turning on both ext-refs and ref-adapt, compared against ext-refs
itself, a small gain is achieved. In PSNR, the bitrate saving gains are
as follows:

lowres: Avg -0.120%; BDRate -0.128%
midres: Avg -0.155%; BDRate -0.128%

Change-Id: I2cfff8a6b7eaa65ef863dbdbc4dd086d3b586f8c

Change-Id: Iad2b526d65865cbcb2119aca21686563ca8e97fd

This ensures we operate at the same precision that Daala uses, which matters
when activity masking is enabled, because of the gain companding.

Metrics from Patchset 4 (which had slightly incorrect rounding):

With activity masking (5 frames only):
av1_pvq_AM_ref_5f@2017-02-07T03:37:53.702Z -> av1_pvq_AM_derf_fix2_coeff_scaling_5f@2017-02-07T00:12:24.427Z

    PSNR |  PSNR Cb |  PSNR Cr | PSNR HVS |   SSIM | MS SSIM | CIEDE 2000
  0.6653 | -12.3177 | -12.1858 |   0.3350 | 4.1013 |  2.0964 |    -4.0539

In particular for Netflix_Crosswalk_1920x1080_60fps_8bit_420_60f.y4m
 -5.0589 | -22.3077 | -21.2188 |  -7.0389 | -3.3715 |-5.7794 |   -13.1891

I.e., it fixes the large regression with AM on this sequence, and
 substantially improves chroma (at a lesser cost to other metrics).

Without activity masking (5 frames only):
av1_pvq_ref_5f@2017-02-07T03:52:51.279Z -> av1_pvq_derf_fix2_coeff_scaling_5f@2017-02-07T00:12:48.873Z

    PSNR | PSNR Cb | PSNR Cr | PSNR HVS |   SSIM | MS SSIM | CIEDE 2000
  0.0989 | -0.0322 | -0.0464 |   0.1883 | 0.0795 |  0.0579 |     0.0923

Change-Id: I46b808b7c8e4733465f8bebc8336dfd5b75783ec

The new_tokenset experiment replaces the unconstrained tokenset with a
multisymbol alphabet in an inventive way.

Tested configurations:
new_tokenset + ec_adapt, new_tokenset, ec_multisymbol

Change-Id: I846ab2e51c2a1dc3f2f9904ed8c47a8e98f853c5

Change-Id: I1050b69045407381d4626b65a0bf6f35957a66f4

RD search and trellis encoding are still sub-optimal.

Change-Id: I233979909118241a0c78761c1d5c2cd6857915e0

BDRATE results:
lowres: -0.880% (up from -0.844%)

Change-Id: I017c0beddcc687148fed33c1e9963e05f1eaf6ea

Change-Id: I378b677cf579441ba0a9014a8a77a1cf3f8b5689

Preparation for merging EOB_TOKEN. The block_zero value
corresponds to the first EOB_TOKEN: other EOB_TOKEN values will
be merged with non-zero values.

Change-Id: I94036783ee240fa916a79c544ecd716a9c24fa59

This commit renames deblocking_across_tiles to loopfilter_across_tiles, 
to get ready for dering and clpf integration.

Change-Id: Id25b051da9b1e5cb92f35a9619662597462d9537

This fixes the following mismatch bugs:
* At the bitstream level, the decoder would not read the delta_qindex
  information for supertx blocks, but the encoder always sent it,
  leading to the encoder and decoder becoming misaligned.
  The delta_qindex information is still required for supertx blocks,
  so change the decoder to read it.
* In addition, the quantizer was not properly adjusted for supertx
  blocks at the decoder. We copy the quantizer setup code from
  non-supertx blocks.

Since this does not change the encoder, it should not have any
quality impact.

Change-Id: I9a0f79c3aa66f2a5a353821e2a6f3b526636e7b4

Change-Id: Ibd4bbceca8218eb94ed0af3c980f8e829a1896de

Zero, one, and two or more coded as one symbol (head).
Remaining tokens coded as a tail symbol.

The pareto CDF distribution is adjusted to cover tokens from
two onwards.

Change-Id: I98b33fab6b9f52690f6ad618ac55e725a97be056

Fixes and turns on the test.

BUG=aomedia:312

Change-Id: I6c7d1970e743ec2b025a798070761d22624e796a

Creates the PVQ_SKIP_TYPE enum to encapsulate the different types of
skipping that can be signaled by PVQ (i.e. skip: AC, DC or both).

There is no impact on the bitstream. However, the decoder will now emit
an internal error if the decoded skip flag is out of range. The
block_skip variable is also renamed to ac_dc_coded as it stores the same
information.

Change-Id: Ib2aadaf99dc1736ea392ae5ed8948c3cdc12da9b

Minor performance changes
0.03% better on lowres
0.01% better on midres

Change-Id: I7a7168f3a2a4d17a03353841a416eff6edf1e241

Now that https://aomedia-review.googlesource.com/#/c/6729/
has been merged, build_intra_predictors_for_interintra() is
now redundant, so replace it by a direct call to
av1_predict_intra_block() and remove the old function.

Reset rect_interintra back to 1.

To do this, we need to make the intra predictor take a
BLOCK_SIZE instead of a TX_SIZE. This is because we need to
be able to predict 32x64 and 64x32 blocks, but there is no
TX_32X64 or TX_64X32.

No effect on output or performance.

Change-Id: I8c185a211c97a85012cc54ec293c785a693608ed

Change-Id: Ieb2922c3df4ef4f8514b8a6df6f9a8fc45ef3cf4

This commit adds a new experiment, Daala's distortion function,
which is designed to better approximate perceptual distortion
in 8x8 pixel blocks.

This experiment is expected to work best with PVQ.

It measures the variance of overlapped 4x4 regions in the 8x8 area,
then uses these variances to scale the MSE of weighted frequency domain
distortion of 8x8 block.

Since AV1 calculates distortion in blocks as small as 4x4, it is not possible to
directly replace the existing distortion functions of AV1,
such as dist_block() and block_rd_txf().
Hence, there has been substantial changes in order to apply
Daala's 8x8 distortion function.
The daala distortion function is applied
after all 4x4 tx blocks in a 8x8 block are encoded (during RDO),
as in below two cases:
1) intra/inter sub8x8 predictions and
2) 4x4 transform with prediction size >= 8.

To enable this experiment, add '--enable-daala-dist' with configure.

TODO: Significant tuning of parameters is required since the function has
originally came from Daala thus most parameters would not work
correctly outside Daala.
The fact that chroma distortion is added to the distortion of AV1's RDO is
also critical since Daala's distortion function is applied to luma only
and chroma continues to use MSE.

Change-Id: If35fdd3aec7efe401f351ba1c99891ad57a3d957

This commit enables the adaptive scan order system support
rectangular trnasform block sizes. It resolves the coding failure
when rect-tx or var-tx are enabled.

BUG=aomedia:143

Change-Id: Ic565284e811e3f7e0ebf2e08fb3748257ce8a049

Change-Id: I4b53dab674390496d8fe7299970c5fb327b5a7be

Change-Id: Id01c785ad48134075c4f6643233413564f0b8fbc

Change-Id: I3d64ec4bbc72143b30a094ece7a6c711d6b479cd

* Change Wiener filter storage to match the format expected
  by the convolve functions

Change-Id: I4d1fb08a13cfc31e69e12c1cb4b2e510c6d8ae30

This will support adapting in each tile.

(https://bugs.chromium.org/p/aomedia/issues/detail?id=71)

Change-Id: I3eced47715749a48f78c4ccf151c4d0b58f36c0d

When segment feature is on, frame level cm->tx_mode can be set to
ONLY_4X4 only if all segments are lossless. Otherwise will cause
bugs when xd->lossless[i] is 0 and xd->lossless[0] is 1.
Also fix the condition of coding tx_type, which should be on when
the qindex of current segment is > 0.

BUG=aomedia:106
BUG=aomedia:104

Change-Id: Ic076083bb78b3b99a6f7d17ec82ee402c64bcc52

At the final round of encoding of each superblock, will go through
each prediction block to check if ncobmc mode is better than non-
overlapped prediction. Note that causal obmc mode is dumped here.

PSNR gain (MOTION_VAR + NCOBMC): -2.845% lowres

Change-Id: Ibe504f7f1882446a08ba426e1e9824bca73bf655

These are under EXT_TX + RECT_TX experiment combo.

Results
=======

Derf Set:
--------
All Intra frames: 1.8% avg improvement (and 1.78% BD-rate improvement)
Video: 0.230% avg improvement (and 0.262% BD-rate improvement)

Objective-1-fast set
--------------------
Video: 0.52 PSNR improvement

Change-Id: I1893465929858e38419f327752dc61c19b96b997

At the edges of the picture only a subset of partitions are legal. Add
new contexts for these borders so they don't distort the probabilities of
the interior of the image where all partitions are legal.

Only include one context for each block size of each border direction
because so few blocks fall into these contexts to begin with.

objective-1-fast:
   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0294 | -0.0911 | -0.2382 |  -0.0481 | -0.0441 | -0.0450 |    -0.0454

derf144: -0.135
lowres: -0.124
midres: -0.076
hdres: -0.078

Change-Id: I909b98eebb7e49273cde90154c8408febe334158

In order to use mvs from a future block in obmc, we first send mbmi
info for the entire superblock, and then call another recursion to
handle the coeffs and recon.

Note: this change is currently not compatible with SUPERTX, later I
will move detoken and recon for supertx to a proper place

Change-Id: I19ab77fa137f53a370e68ea777f70d0306e3e303

In order to reduce the code complexity for handling parameter
coding and recon separately for each 64x64 in non-causal obmc
experiment, we break them down to two steps calling separate
functions, one for params, the other dealing with coefficients
and recon(decoder side).
Note: actually the non-causal prediction can use the original
syntax, but right now in the decoder coeff detoken and recon are
heavily nested.

Change-Id: I72d9c42ab8f38b57850d6b0481551893f1702822

Change-Id: I4a0f0a775362e6e43cd28ed29bf83c912cdc7df5

Change the od_decode_cdf_adapt() function to take an aom_reader
 struct instead of an od_ec_dec struct.
Rename od_decode_cdf_adapt() to aom_decode_cdf_adapt().

Change-Id: I0713d2f56acfea3f67f1b4087c0feee77c2e25cb

Slight improvement in midres and hdres sets of 0.02% and 0.0.9%
respectively.

This is also a better design anyways.

Change-Id: I15b60b8836070a2132641e5b1d8e9f68df426c08