Deb Mukherjee [Mon, 20 May 2013 20:54:39 +0000 (13:54 -0700)]
Refinements on modelcoef expt to reduce storage
Uses more aggrerssive interpolation to reduce storage for the
model tables by almost more than half. Only 48 lists of probs are
stored (as opposed to 128 before), corresponding to ONE_NODE
probabilities of:
1,
3, 7, 11, ..., 115, 119,
127, 135, ..., 247, 255.
Besides, only 1 table is used as opposed to 2 before. So the overall
memory needed for the tables is just 48 * 8 = 384 bytes.
The table currently used is based on a new Pareto distribution with
heavier tail than a generalized Gaussian - which improves results on
derf by about 0.1% over a single table Generaized Gaussian.
Scott LaVarnway [Mon, 20 May 2013 17:03:17 +0000 (13:03 -0400)]
WIP: 4x4 idct/recon merge
This patch eliminates the intermediate diff buffer usage by
combining the short idct and the add residual into one function.
The encoder can use the same code as well.
Jingning Han [Thu, 16 May 2013 05:28:36 +0000 (22:28 -0700)]
Enable bit-stream support to 8x4 and 4x8 partition
The recursive partition type search is enabled down to 4x4, 4x8 and
8x4, followed by the corresponding rate-distortion optimization for
the per-partition encoding mode decisions.
The bit-stream writing/reading synchronized in supporting the
rectangular partition of 8x8 block.
This provides above 1% coding performance gains on derf.
To do next:
1. re-design the rate-distortion loop for inter prediction below 8x8.
2. re-design the rate-distortion loop for intra prediction below 4x4.
3. make the loop-filter aware of rectangular partition of 8x8 block.
4. clean the unused probability models.
5. update default probability values.
Paul Wilkins [Thu, 16 May 2013 15:29:33 +0000 (16:29 +0100)]
Replace default counts with default probs.
Replace vp9_kf_default_bmode_counts structure with
direct default probabilities. The probability structure is
smaller and it removes the need to specify in the bitstream
how to convert the counts to probabilities.
Note that I have concerns still about the size and value of
the large intra mode context. This may cause problems for
HW but it also means we rely heavily on reverse update as
forwards update of a structure this size is problematic. I
intend to review this more generally in the next few days to
see if we can come up with a competitive solution that does
not rely on such a large context.
Jingning Han [Thu, 16 May 2013 20:30:00 +0000 (13:30 -0700)]
Use b_width_log2_ in partition type parse
Replace mi_width/height_log2 with b_width/height_log2 in partition
type parsing at bit-stream writing stage. This allows parsing
resolution at 4x4 block level and makes the 4x4/4x8/8x4 partition
coding consistent with other superblock types.
Yaowu Xu [Tue, 14 May 2013 17:35:11 +0000 (10:35 -0700)]
Removed Q threshold in the usage of ADST
Test on cif set showed small but consistent compression gain for
almost all encodings with overall impact of .08%. The gains average
aournd .12% combined with D63 adst change.
Scott LaVarnway [Thu, 16 May 2013 17:52:15 +0000 (13:52 -0400)]
WIP: 8x8 idct/recon merge
This patch eliminates the intermediate diff buffer usage by
combining the short idct and the add residual into one function.
The encoder can use the same code as well.
Paul Wilkins [Thu, 16 May 2013 10:27:12 +0000 (11:27 +0100)]
New inter mode context.
This patch creates a new inter mode contest that avoids
a dependence on the reconstructed motion vectors from
neighboring blocks. This was a change requested by
a hardware vendor to improve decode performance.
As part of this change I have also made some modifications
to stats output code (under a flag) to allow accumulation of
inter mode context flags over multiple clips
Some further changes will be required to accommodate the
deprecation of the split mv mode over the next few days.
Performance as stands is around -0.25% on derf and
std-hd but up on the YT and YT-HD sets. With further tuning
or some adjustment to the context criteria it should be
possible to make this change broadly neutral.
Dmitry Kovalev [Wed, 15 May 2013 19:37:03 +0000 (12:37 -0700)]
Combining integer and fractional parts of mvs for entropy coding.
Combining encode_nmv_component with encode_nmv_component_fp
and read_nmv_component with read_nmv_component_fp. Bitstream is slightly
changed (only the order of bits), here are the results on test sets:
stdhd: +0.047, yt: -0.038, derf: +0.001, hd: -0.011.
Scott LaVarnway [Wed, 15 May 2013 17:16:02 +0000 (13:16 -0400)]
WIP: 16x16 idct/recon merge
This patch eliminates the intermediate diff buffer usage by
combining the short idct and the add residual into one function.
The encoder can use the same code as well.
Scott LaVarnway [Tue, 14 May 2013 15:58:13 +0000 (11:58 -0400)]
WIP: 32x32 idct/recon merge
This patch eliminates the intermediate diff buffer usage by
combining the short idct and the add residual into one function.
The encoder can use the same code as well.
Jingning Han [Sat, 11 May 2013 00:06:37 +0000 (17:06 -0700)]
Enable recursive partition down to 4x4
This commit allows the rate-distortion optimization recursion
at encoder to go down to 4x4 block size. It deprecates the use
of I4X4_PRED and SPLITMV syntax elements from bit-stream
writing/reading. Will remove the unused probability models in
the next patch.
The partition type search and bit-stream are now capable of
supporting the rectangular partition of 8x8 block, i.e., 8x4
and 4x8. Need to revise the rate-distortion parts to get these
two partition tested in the rd loop.
Yunqing Wang [Sat, 11 May 2013 01:46:43 +0000 (18:46 -0700)]
Do joint motion search iteratively
Allow motion search multiple times iteratively, and break out
the loop if this search couldn't find better motion vectors.
Limit the maximum number of search to 2.
Tests results:
1. stdhd set: 0.311%(overall psnr); 0.346%(ssim).
positive gain on 10 out of 16 clips(best: 2.746% on sunflower;
worst: -0.434% on old_town_cross).
2. derf set: 0.016%(overall psnr); 0.062%(ssim).
positive gain on half of the clips(best: 0.499% on bowing;
worst: -0.387 on city).
Jingning Han [Sat, 11 May 2013 20:33:13 +0000 (13:33 -0700)]
Force bsize for UV in I4X4 and SPLITMV
Use 4x4 block coding for UV components arbitrarily in I4X4_PRED and
SPLITMV coding modes. This is a temporary solution to enable
bit-stream support for recursive partition down to 4x4 block size.
Will separate the functionalities of 4x4 block coding rate-distortion
out from those of superblocks.
Yunqing Wang [Tue, 7 May 2013 16:45:28 +0000 (09:45 -0700)]
Add joint motion search in comp_inter_inter mode(experiment)
In current code, motion vectors got from single prediction mode are used
in compound prediction mode directly. These motion vectors may not give
accurate prediction since they are searched independently. In this patch,
we took Pascal's suggestion, and did joint motion search in compound
prediction mode to find better motion vectors in this situation.
Test results:
Overall PSNR: 0.570%(derf), 0.918%(stdhd);
SSIM: 0.572%(derf), 1.009%(stdhd);
The encoder is a little slower. This can be improved since some c
code is used in motion search.
Jingning Han [Wed, 8 May 2013 21:24:43 +0000 (14:24 -0700)]
Enable recursive partition type search
This commit enables the search for the optimal superblock
partition types in the recursion form. The intention is to
make the optimization process more concise and ready to
support partition down to 4x4 block size next.
projects / libvpx / log