From: Ashish Singh <ashk43...@gmail.com>
Hi, this is vmaf filter. It fuses the scores of previous metrics adm, motion
and vif
using svm algorithm. It's different from libvmaf filter because it has a very
little
external dependency (only one svm model file).
Currently it supports only one model which can be extended later for other
models.
I have added the model file inside libavfilter/data/ so that it can run
successfully.
There is still a bit of work left to do like changing each filter from float
to integer. It's in progress along with SIMD optimizations.
Signed-off-by: Ashish Singh <ashk43...@gmail.com>
---
Changelog | 1 +
doc/filters.texi | 38 ++
libavfilter/Makefile | 1 +
libavfilter/allfilters.c | 1 +
libavfilter/data/vmaf_v0.6.1.pkl.model | 218 ++++++++
libavfilter/vf_vmaf.c | 945 +++++++++++++++++++++++++++++++++
libavfilter/vmaf.h | 138 +++++
7 files changed, 1342 insertions(+)
create mode 100644 libavfilter/data/vmaf_v0.6.1.pkl.model
create mode 100644 libavfilter/vf_vmaf.c
create mode 100644 libavfilter/vmaf.h
diff --git a/Changelog b/Changelog
index 7a6987a..b33f0b2 100644
--- a/Changelog
+++ b/Changelog
@@ -33,6 +33,7 @@ version <next>:
- tlut2 video filter
- floodfill video filter
- pseudocolor video filter
+- vmaf video filter
version 3.3:
- CrystalHD decoder moved to new decode API
diff --git a/doc/filters.texi b/doc/filters.texi
index 3b5a38f..2396d96 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -15375,6 +15375,44 @@ vignette='PI/4+random(1)*PI/50':eval=frame
@end itemize
+@section vmaf
+
+Obtain the VMAF (Video Multi-Method Assessment Fusion)
+score between two input videos.
+
+Both video inputs must have the same resolution and pixel format.
+
+The obtained VMAF score is printed through the logging system.
+
+If no model path is specified it uses the default model:
@code{vmaf_v0.6.1.pkl}.
+
+The filter has following options:
+
+@table @option
+@item model_path
+Set the model path which is to be used for SVM.
+Default value: @code{"vmaf_v0.6.1.pkl.model"}
+
+@item enable_transform
+Enables transform for computing vmaf.
+
+@item pool
+Set the pool method to be used for computing vmaf (mean, min or harmonic mean).
+@end table
+
+On the below examples the input file @file{main.mpg} being processed is
+compared with the reference file @file{ref.mpg}.
+
+For example:
+@example
+ffmpeg -i main.mpg -i ref.mpg -lavfi vmaf -f null -
+@end example
+
+Example with options:
+@example
+ffmpeg -i main.mpg -i ref.mpg -lavfi vmaf="pool=min" -f null -
+@end example
+
@section vstack
Stack input videos vertically.
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 1d92dc1..068b29f 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -327,6 +327,7 @@ OBJS-$(CONFIG_VFLIP_FILTER) += vf_vflip.o
OBJS-$(CONFIG_VIDSTABDETECT_FILTER) += vidstabutils.o
vf_vidstabdetect.o
OBJS-$(CONFIG_VIDSTABTRANSFORM_FILTER) += vidstabutils.o
vf_vidstabtransform.o
OBJS-$(CONFIG_VIGNETTE_FILTER) += vf_vignette.o
+OBJS-$(CONFIG_VMAF_FILTER) += vf_vmaf.o dualinput.o
framesync.o
OBJS-$(CONFIG_VSTACK_FILTER) += vf_stack.o framesync2.o
OBJS-$(CONFIG_W3FDIF_FILTER) += vf_w3fdif.o
OBJS-$(CONFIG_WAVEFORM_FILTER) += vf_waveform.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 8b9b9a4..39c3265 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -338,6 +338,7 @@ static void register_all(void)
REGISTER_FILTER(VIDSTABDETECT, vidstabdetect, vf);
REGISTER_FILTER(VIDSTABTRANSFORM, vidstabtransform, vf);
REGISTER_FILTER(VIGNETTE, vignette, vf);
+ REGISTER_FILTER(VMAF, vmaf, vf);
REGISTER_FILTER(VSTACK, vstack, vf);
REGISTER_FILTER(W3FDIF, w3fdif, vf);
REGISTER_FILTER(WAVEFORM, waveform, vf);
diff --git a/libavfilter/data/vmaf_v0.6.1.pkl.model
b/libavfilter/data/vmaf_v0.6.1.pkl.model
new file mode 100644
index 0000000..fe157df
--- /dev/null
+++ b/libavfilter/data/vmaf_v0.6.1.pkl.model
@@ -0,0 +1,218 @@
+svm_type nu_svr
+kernel_type rbf
+gamma 0.04
+nr_class 2
+total_sv 211
+rho -1.33133
+SV
+-4 1:0.65734273 2:0.34681232 3:0.093755557 4:0.60913934 5:0.69117362
6:0.73495824
+4 1:0.8727433 2:0.49612229 3:0.59146724 4:0.78105663 5:0.84916292 6:0.8882561
+4 1:0.89890005 2:0.49612229 3:0.66823667 4:0.86050887 5:0.90873162
6:0.93335071
+4 1:0.20371751 2:0.49612229 3:0.10534315 4:-1.110223e-16 6:2.220446e-16
+4 1:0.33913836 2:0.49612229 3:0.14024497 4:0.074708413 5:0.10231651
6:0.1259153
+4 1:0.66426757 2:0.49612229 3:0.35268026 4:0.4805681 5:0.59603341 6:0.67408692
+4 1:0.59561632 2:0.49612229 3:0.27561601 4:0.33977371 5:0.4325213 6:0.50244952
+4 1:0.50821444 2:0.49612229 3:0.20276685 4:0.2004308 5:0.25758651 6:0.30054029
+4 1:0.77877298 2:0.49612229 3:0.444392 4:0.61630491 5:0.71210086 6:0.77386496
+4 1:0.71666017 2:0.49612229 3:0.35967401 4:0.47825205 5:0.57045236
6:0.63752441
+4 1:0.64025669 2:0.49612229 3:0.27766156 4:0.33407105 5:0.40732401
6:0.46359154
+4 1:0.88343983 2:0.23066177 3:0.65873851 4:0.86090402 5:0.90661213
6:0.93008753
+4 1:0.90822691 2:0.23066177 3:0.71439481 4:0.90904598 5:0.94146542
6:0.95674338
+-4 1:0.49037399 2:0.23066177 3:0.32329421 4:0.33686197 5:0.39456977
6:0.44944683
+-4 1:0.69044383 2:0.23066177 3:0.43933868 4:0.56327049 5:0.65339511
6:0.71348696
+-4 1:0.62390093 2:0.23066177 3:0.3800888 4:0.44927578 5:0.52327759
6:0.57907725
+4 1:0.81887942 2:0.23066177 3:0.56208506 4:0.76164281 5:0.83176644
6:0.86914911
+4 1:0.77189471 2:0.23066177 3:0.50145055 4:0.66525882 5:0.74327951
6:0.79017822
+4 1:0.71405433 2:0.23066177 3:0.43952897 4:0.55736023 5:0.63319876
6:0.68402869
+4 1:0.92114073 3:0.45198963 4:0.97703695 5:0.9907273 6:0.99510256
+4 1:1 3:0.83319067 4:0.98956086 5:0.99577089 6:0.99784595
+4 4:0.10344019 5:0.34323945 6:0.63855969
+4 1:0.19531482 3:0.034330388 4:0.25480402 5:0.54197045 6:0.78020579
+4 1:0.48394064 3:0.11866359 4:0.58816959 5:0.86435738 6:0.96191842
+4 1:0.47628079 3:0.11185039 4:0.56180003 5:0.83415721 6:0.93617329
+4 1:0.46278632 3:0.10308547 4:0.52247575 5:0.78583924 6:0.89392193
+4 1:0.7038079 3:0.2174879 4:0.84423613 5:0.9662906 6:0.98430594
+4 1:0.69596686 3:0.20657211 4:0.81196884 5:0.94140702 6:0.96680805
+4 1:0.68404358 3:0.19261438 4:0.76066415 5:0.89973293 6:0.93660362
+4 1:0.84073022 2:0.34681232 3:0.22411304 4:0.88845644 5:0.94169671
6:0.96221395
+-4 1:0.33900937 2:0.34681232 3:0.027607294 4:0.40659646 5:0.45456869
6:0.48256597
+-4 1:0.44593129 2:0.34681232 3:0.041939301 4:0.45284872 5:0.5157613
6:0.55335821
+-4 1:0.67301747 2:0.34681232 3:0.11526222 4:0.68549511 5:0.78556255
6:0.83507583
+-4 1:0.62833533 2:0.34681232 3:0.092281981 4:0.61278125 5:0.70626575
6:0.75613977
+-4 1:0.57196879 2:0.34681232 3:0.067548447 4:0.53383404 5:0.61287548
6:0.65468717
+-0.3312466607741135 1:0.75125028 2:0.34681232 3:0.1457048 4:0.75791308
5:0.84155109 6:0.88132116
+-4 1:0.71121936 2:0.34681232 3:0.12095689 4:0.68834617 5:0.77453583
6:0.81892861
+-4 1:0.80269544 2:0.25207203 3:0.3681723 4:0.80658472 5:0.8702283 6:0.90583519
+-4 1:0.86095387 2:0.25207203 3:0.52475418 4:0.85053413 5:0.90454501
6:0.93093678
+-4 1:0.5008963 2:0.25207203 3:0.2005129 4:0.41516485 5:0.45282017 6:0.47396143
+-4 1:0.56977992 2:0.25207203 3:0.21631076 4:0.45848604 5:0.51102137
6:0.53823055
+-4 1:0.72779828 2:0.25207203 3:0.3051639 4:0.67537297 5:0.75767261
6:0.80327187
+-4 1:0.68848569 2:0.25207203 3:0.27393051 4:0.60399854 5:0.68000038
6:0.72275152
+-4 1:0.64121401 2:0.25207203 3:0.23994344 4:0.52538719 5:0.5891732
6:0.62164073
+-4 1:0.76673633 2:0.25207203 3:0.33053889 4:0.73085549 5:0.80341439
6:0.84546456
+-4 1:0.73041172 2:0.25207203 3:0.29691153 4:0.66166141 5:0.73408074
6:0.77757209
+-4 1:0.68529047 2:0.25207203 3:0.26283557 4:0.58611788 5:0.65192525
6:0.69015011
+4 1:0.86902267 2:0.48885268 3:0.5143645 4:0.8587242 5:0.91841685 6:0.94498293
+4 1:0.89266106 2:0.48885268 3:0.55208861 4:0.89938377 5:0.94642982
6:0.96615102
+-4 1:0.42554844 2:0.48885268 3:0.2554221 4:0.36916892 5:0.43100226
6:0.50888404
+-4 1:0.52520274 2:0.48885268 3:0.27824915 4:0.42915458 5:0.50850476
6:0.58585271
+-4 1:0.69357445 2:0.48885268 3:0.35289928 4:0.61359907 5:0.7217863
6:0.78790011
+-4 1:0.64679648 2:0.48885268 3:0.31268451 4:0.5167094 5:0.61224976
6:0.68477529
+4 1:0.80595874 2:0.48885268 3:0.44075432 4:0.7803455 5:0.86328719 6:0.90222545
+-4 1:0.7715192 2:0.48885268 3:0.4012577 4:0.70792536 5:0.80063653 6:0.85083872
+4 1:0.82199966 2:0.20629643 3:0.30562098 4:0.80541317 5:0.89285836
6:0.92907353
+-4 1:0.84774006 2:0.20629643 3:0.36755712 4:0.8681203 5:0.93297792
6:0.95700049
+4 1:0.26631905 2:0.20629643 3:0.076468978 4:0.29833807 5:0.37989948
6:0.4576277
+-4 1:0.65439648 2:0.20629643 3:0.19487894 4:0.63045155 5:0.76931142
6:0.83706632
+4 1:0.55295603 2:0.20629643 3:0.13877412 4:0.4724047 5:0.59295828 6:0.66834832
+4 1:0.75448924 2:0.20629643 3:0.24707248 4:0.72284103 5:0.83178838
6:0.88053503
+4 1:0.83852041 2:0.15600331 3:0.1625414 4:0.81948421 5:0.90185357 6:0.9347395
+4 1:0.85805266 2:0.15600331 3:0.19693206 4:0.86294641 5:0.92990351
6:0.95498998
+-4 1:0.43384835 2:0.15600331 3:0.030541611 4:0.37279112 5:0.4588284
6:0.52004828
+-4 1:0.72588966 2:0.48885268 3:0.35394597 4:0.61189191 5:0.70897304
6:0.77099691
+-4 1:0.65865915 2:0.20629643 3:0.1796405 4:0.56432133 5:0.68049028
6:0.74616621
+-4 1:0.53095193 2:0.15600331 3:0.046271684 4:0.4328793 5:0.5309142
6:0.59282089
+-4 1:0.71891465 2:0.15600331 3:0.11085278 4:0.68794624 5:0.80350923
6:0.85660483
+-4 1:0.68635753 2:0.15600331 3:0.091457045 4:0.60849701 5:0.72282659
6:0.78137183
+-4 1:0.64162333 2:0.15600331 3:0.068820233 4:0.51732819 5:0.62198733
6:0.67977328
+4 1:0.78395225 2:0.15600331 3:0.13401869 4:0.75274384 5:0.8506531 6:0.89321405
+-4 1:0.75276337 2:0.15600331 3:0.11289462 4:0.67598462 5:0.78117168
6:0.83259364
+-4 1:0.71345342 2:0.15600331 3:0.089218917 4:0.58797907 5:0.69284768
6:0.74971699
+4 1:0.93500967 2:0.08765484 3:0.72226864 4:0.93291747 5:0.960644 6:0.97304054
+4 1:0.95150668 2:0.08765484 3:0.77391346 4:0.95596295 5:0.97544784
6:0.98405871
+-4 1:0.48148634 2:0.08765484 3:0.36628046 4:0.45852823 5:0.56005228
6:0.65708595
+-4 1:0.59853216 2:0.08765484 3:0.42071301 4:0.56376512 5:0.66454599 6:0.741236
+-4 1:0.79297271 2:0.08765484 3:0.5597726 4:0.80653689 5:0.88996341
6:0.92691132
+-4 1:0.76798941 2:0.08765484 3:0.52069978 4:0.74484555 5:0.83431246
6:0.87935204
+-4 1:0.73225133 2:0.08765484 3:0.47011786 4:0.66069877 5:0.75226598
6:0.80539407
+-4 1:0.87240592 2:0.08765484 3:0.62680052 4:0.88208508 5:0.93041565
6:0.9505376
+-4 1:0.84834872 2:0.08765484 3:0.58154998 4:0.82429855 5:0.8858516
6:0.91563291
+-4 1:0.84365382 2:0.93973481 3:0.36718425 4:0.81512123 5:0.88887359
6:0.92320992
+-4 1:0.89242364 2:0.93973481 3:0.41336953 4:0.88038833 5:0.93688884
6:0.95992879
+-4 1:0.31373571 2:0.93973481 3:0.18757116 4:0.34864297 5:0.3777168
6:0.38922611
+-4 1:0.42490775 2:0.93973481 3:0.20295859 4:0.39290035 5:0.43632323
6:0.45871216
+-4 1:0.66865444 2:0.93973481 3:0.28594627 4:0.63969879 5:0.73360583
6:0.78380069
+-4 1:0.62642524 2:0.93973481 3:0.26141889 4:0.56602175 5:0.64775366
6:0.69263211
+-4 1:0.57430455 2:0.93973481 3:0.23537634 4:0.48984694 5:0.55363885
6:0.5853905
+-4 1:0.76178555 2:0.93973481 3:0.32205372 4:0.7176044 5:0.80237787
6:0.84588741
+-4 1:0.72282163 2:0.93973481 3:0.29554025 4:0.64471949 5:0.72634443
6:0.77062686
+-4 1:0.67693861 2:0.93973481 3:0.2669659 4:0.56720118 5:0.63868728
6:0.67673331
+4 1:0.86023804 2:0.49739676 3:0.53966638 4:0.77392585 5:0.84784447
6:0.89031641
+1.296591709971377 1:0.31779385 2:0.49739676 3:0.17094319 4:0.12195679
5:0.13277563 6:0.14165413
+4 1:0.68317784 2:0.49739676 3:0.37192301 4:0.52750491 5:0.62426522 6:0.6929947
+4 1:0.55611181 2:0.49739676 3:0.24752355 4:0.28326524 5:0.33261781
6:0.37104424
+4 1:0.7772257 2:0.49739676 3:0.43832146 4:0.63397606 5:0.7240692 6:0.78367237
+4 1:0.66186286 2:0.49739676 3:0.30599867 4:0.39201262 5:0.45927759
6:0.51239284
+4 1:0.94601776 2:0.04579546 3:0.69472114 4:0.97790884 5:0.9891237 6:0.993277
+4 1:0.98838404 2:0.04579546 3:0.90293444 4:0.99181622 5:0.99642641 6:0.9978864
+4 1:0.30006056 2:0.04579546 3:0.31879 4:0.45852885 5:0.59717781 6:0.71487885
+-4 1:0.44902891 2:0.04579546 3:0.35412414 4:0.55926446 5:0.70175505
6:0.79649177
+-4 1:0.69856222 2:0.04579546 3:0.45989947 4:0.82115248 5:0.92520734
6:0.9594384
+-4 1:0.67730161 2:0.04579546 3:0.44400319 4:0.77920819 5:0.88713866
6:0.92903178
+-4 1:0.64419192 2:0.04579546 3:0.42297435 4:0.72390263 5:0.83364665
6:0.88344569
+-4 1:0.80781899 2:0.04579546 3:0.52334234 4:0.88859427 5:0.94013924
6:0.95946903
+-4 1:0.78080761 2:0.04579546 3:0.499439 4:0.84012074 5:0.90229375 6:0.92936693
+4 1:0.97128596 2:0.014623935 3:0.90135809 4:0.99584619 5:0.9970631
6:0.99757649
+4 1:0.99645027 2:0.014623935 3:1 4:1 5:1 6:1
+-4 1:0.5326065 2:0.014623935 3:0.75468972 4:0.76017077 5:0.83753774
6:0.92265059
+-4 1:0.62757004 2:0.014623935 3:0.77708563 4:0.84258654 5:0.91016348
6:0.95440359
+-4 1:0.79306842 2:0.014623935 3:0.78900741 4:0.90386551 5:0.96905764
6:0.98466408
+-4 1:0.77722867 2:0.014623935 3:0.78701408 4:0.89679281 5:0.96056131
6:0.977629
+-4 1:0.75934622 2:0.014623935 3:0.78422805 4:0.88268036 5:0.94383829
6:0.96596858
+-4 1:0.8878718 2:0.014623935 3:0.81445984 4:0.96615706 5:0.98858241
6:0.99176534
+-4 1:0.88211614 2:0.014623935 3:0.81253935 4:0.95982371 5:0.98309178
6:0.9870796
+4 1:0.83805466 2:0.22767235 3:0.31750162 4:0.85145925 5:0.9121085 6:0.93772147
+4 1:0.86620985 2:0.22767235 3:0.35742938 4:0.89821492 5:0.94339974
6:0.96076173
+4 1:0.39289606 2:0.22767235 3:0.12019254 4:0.3951559 5:0.44657802 6:0.46771549
+4 1:0.48692411 2:0.22767235 3:0.13362033 4:0.43434224 5:0.49900609
6:0.53177669
+4 1:0.69743918 2:0.22767235 3:0.2263303 4:0.68859985 5:0.78706365 6:0.83662428
+4 1:0.65237548 2:0.22767235 3:0.19328493 4:0.60107975 5:0.69684945
6:0.74949279
+4 1:0.59461718 2:0.22767235 3:0.15963705 4:0.51010642 5:0.59283393
6:0.63883591
+4 1:0.77302727 2:0.22767235 3:0.26078021 4:0.76359704 5:0.8470807 6:0.8858359
+4 1:0.72953038 2:0.22767235 3:0.22331233 4:0.67735915 5:0.77029889
6:0.81802539
+4 1:0.87210923 2:0.16787772 3:0.69408521 4:0.91495146 5:0.94890261
6:0.96269344
+-4 1:0.81595959 2:0.08765484 3:0.52947327 4:0.7501341 5:0.82294191
6:0.86264385
+4 1:0.72562415 2:0.49739676 3:0.37130724 4:0.51472366 5:0.59961357
6:0.66258291
+-4 1:0.87135693 2:0.014623935 3:0.80905852 4:0.94637428 5:0.97242826
6:0.97946694
+-4 1:0.48910215 2:0.16787772 3:0.49792761 4:0.59161372 5:0.62979552
6:0.64254584
+-4 1:0.5685964 2:0.16787772 3:0.5149767 4:0.63026581 5:0.67890679 6:0.69964851
+-4 1:0.75935478 2:0.16787772 3:0.60695536 4:0.80906778 5:0.87125816
6:0.89810007
+-4 1:0.71788601 2:0.16787772 3:0.57600091 4:0.75310216 5:0.81471966
6:0.84249923
+-4 1:0.66516668 2:0.16787772 3:0.54473368 4:0.69254626 5:0.74796983
6:0.77177867
+4 1:0.81880869 2:0.16787772 3:0.64309172 4:0.86078024 5:0.90892223
6:0.92908907
+-4 1:0.78054558 2:0.16787772 3:0.60849279 4:0.80724494 5:0.86183239
6:0.88618408
+4 1:0.95353512 2:0.055921852 3:0.61526026 4:0.94655706 5:0.97211195
6:0.98210701
+4 1:0.98368527 2:0.055921852 3:0.7405327 4:0.96928567 5:0.9853799 6:0.99080378
+4 1:0.11318821 2:0.055921852 3:0.1590151 4:0.30536689 5:0.48614515
6:0.64344462
+4 1:0.30298819 2:0.055921852 3:0.19401703 4:0.41679982 5:0.61495039
6:0.74140301
+4 1:0.60614412 2:0.055921852 3:0.31791569 4:0.72365433 5:0.88324129
6:0.93484545
+4 1:0.58738733 2:0.055921852 3:0.29301498 4:0.67070014 5:0.83429953
6:0.89348041
+4 1:0.79496816 2:0.055921852 3:0.42192974 4:0.86711004 5:0.94030868
6:0.96084539
+4 1:0.77749763 2:0.055921852 3:0.38714172 4:0.81340799 5:0.90059649
6:0.93006702
+4 1:0.75215882 2:0.055921852 3:0.34721658 4:0.73960747 5:0.84370247
6:0.88485372
+4 1:0.89732805 2:0.58937038 3:0.58823535 4:0.80035053 5:0.86988422
6:0.90533033
+-4 1:0.9228759 2:0.58937038 3:0.65797705 4:0.87169952 5:0.92200942
6:0.94454256
+4 1:0.19504362 2:0.58937038 3:0.21585801 4:0.1754362 5:0.20844015 6:0.23846443
+4 1:0.34425894 2:0.58937038 3:0.24672569 4:0.24188506 5:0.29544562
6:0.33843061
+4 1:0.66407117 2:0.58937038 3:0.40045124 4:0.55415203 5:0.66628031
6:0.73418465
+4 1:0.60780044 2:0.58937038 3:0.34931828 4:0.4519606 5:0.54893247 6:0.61355219
+4 1:0.53476258 2:0.58937038 3:0.29851601 4:0.34826788 5:0.42168642
6:0.47203603
+4 1:0.79195776 2:0.58937038 3:0.47493233 4:0.66775916 5:0.76196439
6:0.81489875
+4 1:0.7415564 2:0.58937038 3:0.41507439 4:0.56413083 5:0.65815516 6:0.7166999
+4 1:0.82021207 2:1 3:0.37381485 4:0.7891612 5:0.87031145 6:0.90944281
+-3.795805084530972 1:0.85903236 2:1 3:0.43235998 4:0.86707094 5:0.92632217
6:0.95151451
+-4 1:0.25243046 2:1 3:0.084027451 4:0.15537936 5:0.17410072 6:0.17212333
+-4 1:0.35643487 2:1 3:0.10644455 4:0.21484368 5:0.25587544 6:0.27527817
+-4 1:0.57605414 2:1 3:0.19031962 4:0.43030863 5:0.5277316 6:0.59069772
+-4 1:0.49071444 2:1 3:0.14452095 4:0.31406915 5:0.38353445 6:0.42653517
+4 1:0.73255545 2:1 3:0.28883701 4:0.65284485 5:0.75623242 6:0.81297442
+0.4082706381617505 1:0.67015395 2:1 3:0.2367756 4:0.5367057 5:0.64063877
6:0.70451767
+-4 1:0.84450653 2:0.083369236 3:0.57279245 4:0.85249389 5:0.91751611
6:0.94621989
+-4 1:0.39559773 2:0.083369236 3:0.28184137 4:0.37025203 5:0.46733936
6:0.53517338
+-4 1:0.70621493 2:0.083369236 3:0.42718441 4:0.69347659 5:0.81124449
6:0.87136343
+-4 1:0.65615861 2:0.083369236 3:0.37833052 4:0.59301482 5:0.71772587
6:0.7905538
+-4 1:0.58837863 2:0.083369236 3:0.33229353 4:0.48675881 5:0.60141743
6:0.67458413
+-4 1:0.77687144 2:0.083369236 3:0.48094343 4:0.76665994 5:0.86191893
6:0.90760934
+-1.966116876631112 1:0.72849768 2:0.083369236 3:0.42082971 4:0.66591147
5:0.77995959 6:0.84260661
+-3.906831378063804 1:0.66320082 2:0.083369236 3:0.36350305 4:0.54888271
5:0.66506794 6:0.73685112
+4 1:0.84500499 2:0.42532178 3:0.43562507 4:0.80721931 5:0.87934044
6:0.91434143
+4 1:0.8874543 2:0.42532178 3:0.50912639 4:0.87959883 5:0.93223488 6:0.95450335
+4 1:0.31032192 2:0.42532178 3:0.18976794 4:0.30662908 5:0.34637104 6:0.3661022
+4 1:0.41026349 2:0.42532178 3:0.20589097 4:0.35241209 5:0.40358156
6:0.42577381
+4 1:0.67552108 2:0.42532178 3:0.30879992 4:0.60375124 5:0.70097073
6:0.75507206
+4 1:0.62772585 2:0.42532178 3:0.27349745 4:0.5196735 5:0.60339149 6:0.65103342
+4 1:0.5741386 2:0.42532178 3:0.24033766 4:0.43855753 5:0.50243186 6:0.53322825
+4 1:0.7629976 2:0.42532178 3:0.35347476 4:0.69239941 5:0.78245146 6:0.83117443
+4 1:0.71746409 2:0.42532178 3:0.31296983 4:0.60525302 5:0.69243388 6:0.7432587
+-4 1:0.73137955 2:0.16787772 3:0.57222383 4:0.74405775 5:0.79993424
6:0.82484891
+4 1:0.67383121 2:0.58937038 3:0.35481019 4:0.45269287 5:0.53578336
6:0.59116487
+4 1:0.5905971 2:1 3:0.18559792 4:0.41535212 5:0.50422336 6:0.56173557
+4 1:0.66157018 2:0.42532178 3:0.27479904 4:0.51802649 5:0.59270541
6:0.63560969
+-4 1:0.66827754 2:0.54342577 3:0.18169339 4:0.50290989 5:0.59875259
6:0.65332628
+4 1:0.85027066 2:0.20820673 3:0.40997978 4:0.82462749 5:0.89794736
6:0.93142825
+4 1:0.87892054 2:0.20820673 3:0.45891267 4:0.87823329 5:0.93535353
6:0.95883927
+4 1:0.3986268 2:0.20820673 3:0.17753958 4:0.33495583 5:0.39777832 6:0.44399359
+-4 1:0.48997993 2:0.20820673 3:0.20172681 4:0.39715881 5:0.47368229
6:0.52781628
+4 1:0.7022939 2:0.20820673 3:0.31094767 4:0.6676259 5:0.77726116 6:0.83518027
+-4 1:0.65773092 2:0.20820673 3:0.27420721 4:0.57889989 5:0.68485118
6:0.74837036
+0.2951376518668717 1:0.60031736 2:0.20820673 3:0.23419121 4:0.48018865
5:0.57200972 6:0.63197473
+4 1:0.77623676 2:0.20820673 3:0.3510016 4:0.74206651 5:0.83508543 6:0.88101902
+4 1:0.73562396 2:0.20820673 3:0.31004997 4:0.6557112 5:0.75585014 6:0.81164989
+-4 1:0.67923081 2:0.20820673 3:0.26679137 4:0.55816547 5:0.65579282
6:0.71593631
+4 1:0.83968539 2:0.54342577 3:0.32439292 4:0.78747769 5:0.87303614
6:0.91271252
+4 1:0.86656342 2:0.54342577 3:0.37898741 4:0.85252726 5:0.92049615
6:0.94848246
+-4 1:0.42728303 2:0.54342577 3:0.10123262 4:0.31581962 5:0.38571265
6:0.42827036
+-4 1:0.63194526 2:0.54342577 3:0.18169045 4:0.51611903 5:0.62179755
6:0.68216176
+4 1:0.56954706 2:0.54342577 3:0.14271477 4:0.41491191 5:0.50173488
6:0.55220392
+4 1:0.76753176 2:0.54342577 3:0.26295318 4:0.6905031 5:0.79291823 6:0.84469464
+-4 1:0.72348649 2:0.54342577 3:0.22334634 4:0.60145902 5:0.70573225
6:0.76318544
+4 1:0.83584492 2:0.047285912 3:0.53826775 4:0.933335 5:0.95948954 6:0.96870909
+4 1:0.85530855 2:0.047285912 3:0.55323777 4:0.95113339 5:0.97249918
6:0.9795177
+-4 1:0.53835734 2:0.047285912 3:0.41965074 4:0.71632669 5:0.73953043
6:0.73487553
+-4 1:0.59175144 2:0.047285912 3:0.43113594 4:0.74141738 5:0.76929188
6:0.77018949
+-4 1:0.75962366 2:0.047285912 3:0.49613729 4:0.87838146 5:0.91688438
6:0.93150362
+-4 1:0.72043129 2:0.047285912 3:0.47217411 4:0.83138845 5:0.8704229
6:0.88419439
+-4 1:0.67287449 2:0.047285912 3:0.44652268 4:0.77691812 5:0.81043483
6:0.8177009
+-4 1:0.8023177 2:0.047285912 3:0.51559706 4:0.90512389 5:0.93743101
6:0.9492968
+-4 1:0.76751376 2:0.047285912 3:0.49225957 4:0.86357299 5:0.89948127
6:0.91221155
+-4 1:0.72124785 2:0.047285912 3:0.46606653 4:0.81323145 5:0.84847474
6:0.85892657
diff --git a/libavfilter/vf_vmaf.c b/libavfilter/vf_vmaf.c
new file mode 100644
index 0000000..4cb93eb
--- /dev/null
+++ b/libavfilter/vf_vmaf.c
@@ -0,0 +1,945 @@
+/*
+ * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com>
+ * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Calculate the VMAF between two input videos.
+ */
+
+#include "libavutil/avstring.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "avfilter.h"
+#include "dualinput.h"
+#include "drawutils.h"
+#include "formats.h"
+#include "internal.h"
+#include "video.h"
+#include "adm.h"
+#include "vmaf_motion.h"
+#include "vif.h"
+#include "vmaf.h"
+
+typedef struct VMAFContext {
+ const AVClass *class;
+ FFDualInputContext dinput;
+ const AVPixFmtDescriptor *desc;
+ int width;
+ int height;
+ uint8_t called;
+ double score;
+ double scores[8];
+ double score_num;
+ double score_den;
+ int conv_filter[5];
+ float *ref_data;
+ float *main_data;
+ float *adm_data_buf;
+ float *adm_temp_lo;
+ float *adm_temp_hi;
+ uint16_t *prev_blur_data;
+ uint16_t *blur_data;
+ uint16_t *temp_data;
+ float *vif_data_buf;
+ float *vif_temp;
+ double prev_motion_score;
+ double vmaf_score;
+ uint64_t nb_frames;
+ char *model_path;
+ int enable_transform;
+ char *pool;
+ svm_model *svm_model_ptr;
+ svm_node* nodes;
+ void (*pool_method)(double *score, double curr);
+ double prediction;
+} VMAFContext;
+
+#define OFFSET(x) offsetof(VMAFContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+
+static const AVOption vmaf_options[] = {
+ {"model_path", "Set the model to be used for computing vmaf.",
OFFSET(model_path), AV_OPT_TYPE_STRING,
{.str="libavfilter/data/vmaf_v0.6.1.pkl.model"}, 0, 1, FLAGS},
+ {"enable_transform", "Enables transform for computing vmaf.",
OFFSET(enable_transform), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
+ {"pool", "Set the pool method to be used for computing vmaf.",
OFFSET(pool), AV_OPT_TYPE_STRING, {.str="mean"}, 0, 1, FLAGS},
+ { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(vmaf);
+
+#define MAX_ALIGN 32
+#define ALIGN_CEIL(x) ((x) + ((x) % MAX_ALIGN ? MAX_ALIGN - (x) % MAX_ALIGN :
0))
+
+enum { C_SVC, NU_SVC, ONE_CLASS, EPSILON_SVR, NU_SVR }; /** svm_type */
+enum { LINEAR, POLY, RBF, SIGMOID, PRECOMPUTED }; /** kernel_type */
+
+#define swap(type, x, y) { type t=x; x=y; y=t; }
+
+static inline double power(double base, int times)
+{
+ double tmp = base, ret = 1.0;
+
+ for(int t = times; t > 0; t /= 2) {
+ if(t % 2 == 1) {
+ ret *= tmp;
+ }
+ tmp = tmp * tmp;
+ }
+ return ret;
+}
+
+static double dot(const svm_node *px, const svm_node *py)
+{
+ double sum = 0;
+ while(px->index != -1 && py->index != -1) {
+ if(px->index == py->index) {
+ sum += px->value * py->value;
+ px++;
+ py++;
+ } else {
+ if(px->index > py->index) {
+ py++;
+ } else {
+ px++;
+ }
+ }
+ }
+ return sum;
+}
+
+static double k_function(const svm_node *x, const svm_node *y,
+ const svm_parameter *param)
+{
+ switch(param->kernel_type) {
+ case LINEAR:
+ return dot(x, y);
+ case POLY:
+ return power(param->gamma * dot(x, y) + param->coef0,
param->degree);
+ case RBF: {
+ double sum = 0;
+ while(x->index != -1 && y->index !=-1) {
+ if(x->index == y->index) {
+ double d = x->value - y->value;
+ sum += d * d;
+ x++;
+ y++;
+ } else {
+ if(x->index > y->index) {
+ sum += y->value * y->value;
+ y++;
+ } else {
+ sum += x->value * x->value;
+ x++;
+ }
+ }
+ }
+
+ while(x->index != -1) {
+ sum += x->value * x->value;
+ x++;
+ }
+
+ while(y->index != -1) {
+ sum += y->value * y->value;
+ y++;
+ }
+
+ return exp(-param->gamma * sum);
+ }
+ case SIGMOID:
+ return tanh(param->gamma * dot(x, y) + param->coef0);
+ case PRECOMPUTED:
+ return x[(int)(y->value)].value;
+ default:
+ return 0;
+ }
+}
+
+#define INF HUGE_VAL
+#define TAU 1e-12
+#define Malloc(type,n) (type *)malloc((n)*sizeof(type))
+
+static double svm_predict_values(const svm_model *model, const svm_node *x,
double* dec_values)
+{
+ int i, j;
+ if(model->param.svm_type == ONE_CLASS ||
+ model->param.svm_type == EPSILON_SVR ||
+ model->param.svm_type == NU_SVR) {
+ double *sv_coef = model->sv_coef[0];
+ double sum = 0;
+ for(i = 0; i < model->l; i++) {
+ sum += sv_coef[i] * k_function(x, model->SV[i], &model->param);
+ }
+ sum -= model->rho[0];
+ *dec_values = sum;
+
+ if(model->param.svm_type == ONE_CLASS) {
+ return (sum > 0) ? 1 : -1;
+ } else {
+ return sum;
+ }
+ } else {
+ int nr_class = model->nr_class;
+ int l = model->l;
+ int *start;
+ int *vote;
+ int p;
+ int vote_max_idx;
+ double *kvalue = Malloc(double,l);
+ for(i = 0; i < l; i++) {
+ kvalue[i] = k_function(x, model->SV[i], &model->param);
+ }
+
+ start = Malloc(int,nr_class);
+ start[0] = 0;
+ for(i = 1; i < nr_class; i++) {
+ start[i] = start[i - 1] + model->nSV[i - 1];
+ }
+
+ vote = Malloc(int,nr_class);
+ for(i = 0; i < nr_class; i++) {
+ vote[i] = 0;
+ }
+
+ p=0;
+ for(i = 0; i < nr_class; i++) {
+ for(j = i + 1; j < nr_class; j++) {
+ double sum = 0;
+ int si = start[i];
+ int sj = start[j];
+ int ci = model->nSV[i];
+ int cj = model->nSV[j];
+
+ int k;
+ double *coef1 = model->sv_coef[j - 1];
+ double *coef2 = model->sv_coef[i];
+ for(k = 0; k < ci; k++) {
+ sum += coef1[si + k] * kvalue[si + k];
+ }
+ for(k = 0; k < cj; k++) {
+ sum += coef2[sj + k] * kvalue[sj + k];
+ }
+ sum -= model->rho[p];
+ dec_values[p] = sum;
+
+ if(dec_values[p] > 0) {
+ vote[i]++;
+ } else {
+ vote[j]++;
+ }
+ p++;
+ }
+ }
+
+ vote_max_idx = 0;
+ for(i = 1; i < nr_class; i++) {
+ if(vote[i] > vote[vote_max_idx]) {
+ vote_max_idx = i;
+ }
+ }
+
+ av_free(kvalue);
+ av_free(start);
+ av_free(vote);
+ return model->label[vote_max_idx];
+ }
+}
+
+static double svm_predict(const svm_model *model, const svm_node *x)
+{
+ int nr_class = model->nr_class;
+ double *dec_values;
+ double pred_result;
+ if(model->param.svm_type == ONE_CLASS ||
+ model->param.svm_type == EPSILON_SVR ||
+ model->param.svm_type == NU_SVR) {
+ dec_values = Malloc(double, 1);
+ } else {
+ dec_values = Malloc(double, nr_class * (nr_class - 1) / 2);
+ }
+ pred_result = svm_predict_values(model, x, dec_values);
+ av_free(dec_values);
+ return pred_result;
+}
+
+static const char *svm_type_table[] =
+{
+ "c_svc","nu_svc","one_class","epsilon_svr","nu_svr",NULL
+};
+
+static const char *kernel_type_table[]=
+{
+ "linear","polynomial","rbf","sigmoid","precomputed",NULL
+};
+
+static char *line = NULL;
+static int max_line_len;
+
+static char* readline(FILE *input)
+{
+ int len;
+
+ if(fgets(line,max_line_len,input) == NULL) {
+ return NULL;
+ }
+
+ while(strrchr(line, '\n') == NULL) {
+ max_line_len *= 2;
+ line = (char *) realloc(line, max_line_len);
+ len = (int) strlen(line);
+ if(fgets(line+len, max_line_len-len, input) == NULL) {
+ break;
+ }
+ }
+ return line;
+}
+
+/** FSCANF helps to handle fscanf failures.
+ * Its do-while block avoids the ambiguity when
+ * if (...)
+ * FSCANF();
+ * is used
+ */
+#define FSCANF(_stream, _format, _var) do{ if (fscanf(_stream, _format, _var)
!= 1) return 0; }while(0)
+static int read_model_header(FILE *fp, svm_model* model, AVFilterContext *ctx)
+{
+ svm_parameter* param = &model->param;
+ char cmd[81];
+ int i;
+ while(1) {
+ FSCANF(fp, "%80s", cmd);
+
+ if(av_strcasecmp(cmd, "svm_type") == 0) {
+ FSCANF(fp, "%80s", cmd);
+ for(i = 0; svm_type_table[i]; i++) {
+ if(av_strcasecmp(svm_type_table[i], cmd) == 0) {
+ param->svm_type = i;
+ break;
+ }
+ }
+ if(svm_type_table[i] == NULL) {
+ av_log(ctx, AV_LOG_ERROR, "unknown svm type.\n");
+ return 0;
+ }
+ } else if(av_strcasecmp(cmd, "kernel_type") == 0) {
+ FSCANF(fp, "%80s", cmd);
+ for(i = 0; kernel_type_table[i]; i++) {
+ if(av_strcasecmp(kernel_type_table[i], cmd) == 0) {
+ param->kernel_type = i;
+ break;
+ }
+ }
+ if(kernel_type_table[i] == NULL) {
+ av_log(ctx, AV_LOG_ERROR, "unknown kernel function.\n");
+ return 0;
+ }
+ } else if(av_strcasecmp(cmd, "degree") == 0) {
+ FSCANF(fp, "%d", ¶m->degree);
+ } else if(av_strcasecmp(cmd, "gamma") == 0) {
+ FSCANF(fp, "%lf", ¶m->gamma);
+ } else if(av_strcasecmp(cmd, "coef0") == 0) {
+ FSCANF(fp, "%lf", ¶m->coef0);
+ } else if(av_strcasecmp(cmd, "nr_class") == 0) {
+ FSCANF(fp,"%d",&model->nr_class);
+ } else if(av_strcasecmp(cmd, "total_sv") == 0) {
+ FSCANF(fp, "%d", &model->l);
+ } else if(av_strcasecmp(cmd, "rho")==0) {
+ int n = model->nr_class * (model->nr_class-1) / 2;
+ model->rho = Malloc(double, n);
+ for(i = 0; i < n; i++) {
+ FSCANF(fp, "%lf", &model->rho[i]);
+ }
+ } else if(av_strcasecmp(cmd, "label") == 0) {
+ int n = model->nr_class;
+ model->label = Malloc(int, n);
+ for(i = 0; i < n; i++) {
+ FSCANF(fp, "%d", &model->label[i]);
+ }
+ } else if(av_strcasecmp(cmd, "probA") == 0) {
+ int n = model->nr_class * (model->nr_class - 1) / 2;
+ model->probA = Malloc(double, n);
+ for(i = 0;i < n; i++) {
+ FSCANF(fp, "%lf", &model->probA[i]);
+ }
+ } else if(av_strcasecmp(cmd, "probB") == 0) {
+ int n = model->nr_class * (model->nr_class - 1) / 2;
+ model->probB = Malloc(double,n);
+ for(i = 0; i < n; i++) {
+ FSCANF(fp, "%lf", &model->probB[i]);
+ }
+ } else if(av_strcasecmp(cmd, "nr_sv") == 0) {
+ int n = model->nr_class;
+ model->nSV = Malloc(int, n);
+ for(i = 0; i < n; i++) {
+ FSCANF(fp, "%d", &model->nSV[i]);
+ }
+ } else if(av_strcasecmp(cmd, "SV") == 0) {
+ while(1) {
+ int c = getc(fp);
+ if(c == EOF || c == '\n') {
+ break;
+ }
+ }
+ break;
+ } else {
+ av_log(ctx, AV_LOG_ERROR, "unknown text in model file: [%s]\n",
cmd);
+ return 0;
+ }
+ }
+
+ return 1;
+
+}
+
+static svm_model *svm_load_model(const char *model_file_name, AVFilterContext
*ctx)
+{
+ FILE *fp = fopen(model_file_name, "rb");
+ int i, j, k, l, m;
+ char *p, *endptr, *idx, *val;
+ svm_model *model;
+
+ int elements;
+ long pos;
+ svm_node *x_space;
+
+ if(fp == NULL) {
+ return NULL;
+ }
+
+ /** read parameters */
+ model = Malloc(svm_model,1);
+ model->rho = NULL;
+ model->probA = NULL;
+ model->probB = NULL;
+ model->sv_indices = NULL;
+ model->label = NULL;
+ model->nSV = NULL;
+
+ /** read header */
+ if (!read_model_header(fp, model, ctx)) {
+ av_log(ctx, AV_LOG_ERROR, "ERROR: fscanf failed to read model\n");
+ av_free(model->rho);
+ av_free(model->label);
+ av_free(model->nSV);
+ av_free(model);
+ return NULL;
+ }
+
+ /** read sv_coef and SV */
+ elements = 0;
+ pos = ftell(fp);
+
+ max_line_len = 1024;
+ line = Malloc(char, max_line_len);
+
+ while(readline(fp) != NULL) {
+ p = strtok(line, ":");
+ while(1) {
+ p = strtok(NULL, ":");
+ if(p == NULL) {
+ break;
+ }
+ elements++;
+ }
+ }
+ elements += model->l;
+
+ fseek(fp, pos, SEEK_SET);
+
+ m = model->nr_class - 1;
+ l = model->l;
+ model->sv_coef = Malloc(double *,m);
+ for(i = 0; i < m; i++) {
+ model->sv_coef[i] = Malloc(double,l);
+ }
+ model->SV = Malloc(svm_node*, l);
+ x_space = NULL;
+ if(l > 0) {
+ x_space = Malloc(svm_node, elements);
+ }
+
+ j=0;
+ for(i = 0; i < l; i++) {
+ readline(fp);
+ model->SV[i] = &x_space[j];
+
+ p = strtok(line, " \t");
+ model->sv_coef[0][i] = strtod(p, &endptr);
+ for(k = 1; k < m; k++) {
+ p = strtok(NULL, " \t");
+ model->sv_coef[k][i] = strtod(p, &endptr);
+ }
+
+ while(1) {
+ idx = strtok(NULL, ":");
+ val = strtok(NULL, " \t");
+
+ if(val == NULL) {
+ break;
+ }
+ x_space[j].index = (int) strtol(idx, &endptr, 10);
+ x_space[j].value = strtod(val, &endptr);
+
+ j++;
+ }
+ x_space[j++].index = -1;
+ }
+ av_free(line);
+
+ if (ferror(fp) != 0 || fclose(fp) != 0) {
+ return NULL;
+ }
+
+ model->free_sv = 1;
+ return model;
+}
+
+static void svm_free_model_content(svm_model* model_ptr)
+{
+ int i;
+ if(model_ptr->free_sv && model_ptr->l > 0 && model_ptr->SV != NULL) {
+ av_free((void *) (model_ptr->SV[0]));
+ }
+ if(model_ptr->sv_coef) {
+ for(i = 0; i < model_ptr->nr_class - 1; i++) {
+ av_free(model_ptr->sv_coef[i]);
+ }
+ }
+
+ av_free(model_ptr->SV);
+ model_ptr->SV = NULL;
+
+ av_free(model_ptr->sv_coef);
+ model_ptr->sv_coef = NULL;
+
+ av_free(model_ptr->rho);
+ model_ptr->rho = NULL;
+
+ av_free(model_ptr->label);
+ model_ptr->label= NULL;
+
+ av_free(model_ptr->probA);
+ model_ptr->probA = NULL;
+
+ av_free(model_ptr->probB);
+ model_ptr->probB= NULL;
+
+ av_free(model_ptr->sv_indices);
+ model_ptr->sv_indices = NULL;
+
+ av_free(model_ptr->nSV);
+ model_ptr->nSV = NULL;
+}
+
+static void svm_free_and_destroy_model(svm_model** model_ptr_ptr)
+{
+ if(model_ptr_ptr != NULL && *model_ptr_ptr != NULL) {
+ svm_free_model_content(*model_ptr_ptr);
+ av_free(*model_ptr_ptr);
+ *model_ptr_ptr = NULL;
+ }
+}
+
+static void mean(double *score, double curr)
+{
+ *score += curr;
+}
+
+static void min(double *score, double curr)
+{
+ *score = FFMIN(*score, curr);
+}
+
+static void harmonic_mean(double *score, double curr)
+{
+ *score += 1.0 / (curr + 1.0);
+}
+
+#define offset_fn(type, bits) \
+ static void offset_##bits##bit(VMAFContext *s, const AVFrame *ref, AVFrame
*main, int stride) \
+{ \
+ int w = s->width; \
+ int h = s->height; \
+ int i,j; \
+ \
+ ptrdiff_t ref_stride = ref->linesize[0]; \
+ ptrdiff_t main_stride = main->linesize[0]; \
+ \
+ const type *ref_ptr = (const type *) ref->data[0]; \
+ const type *main_ptr = (const type *) main->data[0]; \
+ \
+ float *ref_ptr_data = s->ref_data; \
+ float *main_ptr_data = s->main_data; \
+ \
+ for(i = 0; i < h; i++) { \
+ for(j = 0; j < w; j++) { \
+ ref_ptr_data[j] = (float) ref_ptr[j]; \
+ main_ptr_data[j] = (float) main_ptr[j]; \
+ } \
+ ref_ptr += ref_stride / sizeof(type); \
+ ref_ptr_data += stride / sizeof(float); \
+ main_ptr += main_stride / sizeof(type); \
+ main_ptr_data += stride / sizeof(float); \
+ } \
+}
+
+offset_fn(uint8_t, 8);
+offset_fn(uint16_t, 10);
+
+static int compute_vmaf(const AVFrame *ref, AVFrame *main, void *ctx)
+{
+ VMAFContext *s = (VMAFContext *) ctx;
+
+ size_t motion_data_sz;
+ int i,j;
+ ptrdiff_t ref_stride;
+ ptrdiff_t ref_px_stride;
+ ptrdiff_t stride;
+ ptrdiff_t motion_stride;
+ ptrdiff_t motion_px_stride;
+ int w = s->width;
+ int h = s->height;
+
+ ref_stride = ref->linesize[0];
+
+ stride = ALIGN_CEIL(w * sizeof(float));
+ motion_stride = ALIGN_CEIL(w * sizeof(uint16_t));
+ motion_px_stride = motion_stride / sizeof(uint16_t);
+
+ /** Offset ref and main pixel by OPT_RANGE_PIXEL_OFFSET */
+ if (s->desc->comp[0].depth <= 8) {
+ offset_8bit(s, ref, main, stride);
+ } else {
+ offset_10bit(s, ref, main, stride);
+ }
+
+ motion_data_sz = (size_t)motion_stride * s->height;
+
+ compute_adm2(s->ref_data, s->main_data, w, h, stride, stride, &s->score,
+ &s->score_num, &s->score_den, s->scores, s->adm_data_buf,
+ s->adm_temp_lo, s->adm_temp_hi);
+ s->nodes[0].index = 1;
+ s->nodes[0].value = (double)(slopes[1]) * (double)(s->score_num /
s->score_den) + (double)(intercepts[1]);
+
+ if (s->desc->comp[0].depth <= 8) {
+ ref_px_stride = ref_stride / sizeof(uint8_t);
+ convolution_f32(s->conv_filter, 5, (const uint8_t *) ref->data[0],
+ s->blur_data, s->temp_data, s->width, s->height,
+ ref_px_stride, motion_px_stride, 8);
+ } else {
+ ref_px_stride = ref_stride / sizeof(uint16_t);
+ convolution_f32(s->conv_filter, 5, (const uint16_t *) ref->data[0],
+ s->blur_data, s->temp_data, s->width, s->height,
+ ref_px_stride, motion_px_stride, 10);
+ }
+
+ if(!s->nb_frames) {
+ s->score = 0.0;
+ } else {
+ compute_vmafmotion(s->prev_blur_data, s->blur_data, s->width,
s->height,
+ motion_stride, motion_stride, &s->score);
+ }
+
+ memcpy(s->prev_blur_data, s->blur_data, motion_data_sz);
+
+ s->nodes[1].index = 2;
+ s->nodes[1].value = (double)(slopes[2]) *
(double)FFMIN(s->prev_motion_score, s->score) + (double)(intercepts[2]);
+ s->prev_motion_score = s->score;
+
+ compute_vif2(s->ref_data, s->main_data, w, h, stride, stride, &s->score,
+ &s->score_num, &s->score_den, s->scores, s->vif_data_buf,
+ s->vif_temp);
+
+ j = 0;
+ for(i = 0; j < 4; i += 2) {
+ s->nodes[j+2].index = j+3;
+ s->nodes[j+2].value = (double)(slopes[j+3]) * (double)((s->scores[i])
/ (s->scores[i+1])) + (double)(intercepts[j+3]);
+ j++;
+ }
+
+ s->prediction = svm_predict(s->svm_model_ptr, s->nodes);
+
+ if (!av_strcasecmp(norm_type, "linear_rescale")) {
+ /** denormalize */
+ s->prediction = (s->prediction - (double)(intercepts[0])) /
(double)(slopes[0]);
+ }
+
+ /** score transform */
+ if (s->enable_transform) {
+ double value = 0.0;
+
+ /** quadratic transform */
+ value += (double)(score_transform[0]);
+ value += (double)(score_transform[1]) * s->prediction;
+ value += (double)(score_transform[2]) * s->prediction * s->prediction;
+
+ /** rectification */
+ if (value < s->prediction) {
+ value = s->prediction;
+ }
+
+ s->prediction = value;
+ }
+
+ s->pool_method(&s->vmaf_score, s->prediction);
+ return 0;
+}
+
+static AVFrame *do_vmaf(AVFilterContext *ctx, AVFrame *main, const AVFrame
*ref)
+{
+ VMAFContext *s = ctx->priv;
+
+ compute_vmaf(ref, main, s);
+
+ s->nb_frames++;
+
+ return main;
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+ VMAFContext *s = ctx->priv;
+
+ if(!s->called) {
+ int i;
+ for(i = 0; i < 5; i++) {
+ s->conv_filter[i] = lrint(FILTER_5[i] * (1 << N));
+ }
+
+ s->svm_model_ptr = svm_load_model(s->model_path, ctx);
+ s->nodes = (svm_node *) av_malloc(sizeof(svm_node) * (6 + 1));
+ s->nodes[6].index = -1;
+ if(!av_strcasecmp(s->pool, "mean")) {
+ s->pool_method = mean;
+ } else if(!av_strcasecmp(s->pool, "min")) {
+ s->vmaf_score = INT_MAX;
+ s->pool_method = min;
+ } else if(!av_strcasecmp(s->pool, "harmonic")) {
+ s->pool_method = harmonic_mean;
+ }
+ }
+
+ s->called = 1;
+ s->dinput.process = do_vmaf;
+
+ return 0;
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+ static const enum AVPixelFormat pix_fmts[] = {
+ AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P,
+ AV_PIX_FMT_YUV444P10LE, AV_PIX_FMT_YUV422P10LE, AV_PIX_FMT_YUV420P10LE,
+ AV_PIX_FMT_NONE
+ };
+
+ AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
+ if (!fmts_list)
+ return AVERROR(ENOMEM);
+ return ff_set_common_formats(ctx, fmts_list);
+}
+
+static int config_input_ref(AVFilterLink *inlink)
+{
+ AVFilterContext *ctx = inlink->dst;
+ VMAFContext *s = ctx->priv;
+ ptrdiff_t stride;
+ size_t data_sz;
+ ptrdiff_t adm_buf_stride;
+ size_t adm_buf_sz;
+ ptrdiff_t vif_buf_stride;
+ size_t vif_buf_sz;
+
+ if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
+ ctx->inputs[0]->h != ctx->inputs[1]->h) {
+ av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be
same.\n");
+ return AVERROR(EINVAL);
+ }
+ if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
+ av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
+ return AVERROR(EINVAL);
+ }
+
+ s->desc = av_pix_fmt_desc_get(inlink->format);
+ s->width = ctx->inputs[0]->w;
+ s->height = ctx->inputs[0]->h;
+
+
+ stride = ALIGN_CEIL(s->width * sizeof(float));
+ data_sz = (size_t)stride * s->height;
+
+ if (!(s->ref_data = av_malloc(data_sz))) {
+ return AVERROR(ENOMEM);
+ }
+
+ if (!(s->main_data = av_malloc(data_sz))) {
+ return AVERROR(ENOMEM);
+ }
+
+ adm_buf_stride = ALIGN_CEIL(((s->width + 1) / 2) * sizeof(float));
+ adm_buf_sz = (size_t)adm_buf_stride * ((s->height + 1) / 2);
+
+ if (SIZE_MAX / adm_buf_sz < 35) {
+ av_log(ctx, AV_LOG_ERROR, "error: SIZE_MAX / buf_sz_one < 35.\n");
+ return AVERROR(EINVAL);
+ }
+
+ if (!(s->adm_data_buf = av_malloc(adm_buf_sz * 35))) {
+ return AVERROR(ENOMEM);
+ }
+
+ if (!(s->adm_temp_lo = av_malloc(stride))) {
+ return AVERROR(ENOMEM);
+ }
+ if (!(s->adm_temp_hi = av_malloc(stride))) {
+ return AVERROR(ENOMEM);
+ }
+
+ stride = ALIGN_CEIL(s->width * sizeof(uint16_t));
+ data_sz = (size_t)stride * s->height;
+
+ if (!(s->prev_blur_data = av_mallocz(data_sz))) {
+ return AVERROR(ENOMEM);
+ }
+
+ if (!(s->blur_data = av_mallocz(data_sz))) {
+ return AVERROR(ENOMEM);
+ }
+
+ if (!(s->temp_data = av_mallocz(data_sz))) {
+ return AVERROR(ENOMEM);
+ }
+
+ vif_buf_stride = ALIGN_CEIL(s->width * sizeof(float));
+ vif_buf_sz = (size_t)vif_buf_stride * s->height;
+
+ if (SIZE_MAX / data_sz < 15) {
+ av_log(ctx, AV_LOG_ERROR, "error: SIZE_MAX / buf_sz < 15\n");
+ return AVERROR(EINVAL);
+ }
+
+ if (!(s->vif_data_buf = av_malloc(vif_buf_sz * 16))) {
+ return AVERROR(ENOMEM);
+ }
+
+ if (!(s->vif_temp = av_malloc(s->width * sizeof(float)))) {
+ return AVERROR(ENOMEM);
+ }
+
+ return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+ AVFilterContext *ctx = outlink->src;
+ VMAFContext *s = ctx->priv;
+ AVFilterLink *mainlink = ctx->inputs[0];
+ int ret;
+
+ outlink->w = mainlink->w;
+ outlink->h = mainlink->h;
+ outlink->time_base = mainlink->time_base;
+ outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
+ outlink->frame_rate = mainlink->frame_rate;
+ if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0)
+ return ret;
+
+ return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
+{
+ VMAFContext *s = inlink->dst->priv;
+ return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+ VMAFContext *s = outlink->src->priv;
+ return ff_dualinput_request_frame(&s->dinput, outlink);
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+ VMAFContext *s = ctx->priv;
+
+ if (s->nb_frames > 0) {
+ if(!av_strcasecmp(s->pool, "mean")) {
+ s->vmaf_score = s->vmaf_score / s->nb_frames;
+ } else if(!av_strcasecmp(s->pool, "min")) {
+ s->vmaf_score = s->vmaf_score;
+ } else if(!av_strcasecmp(s->pool, "harmonic")) {
+ s->vmaf_score = 1.0 / (s->vmaf_score / s->nb_frames) - 1.0;
+ }
+
+ av_log(ctx, AV_LOG_INFO, "VMAF Score: %.3f\n", s->vmaf_score);
+
+ svm_free_and_destroy_model((svm_model **)&s->svm_model_ptr);
+
+ av_free(s->ref_data);
+ av_free(s->main_data);
+ av_free(s->adm_data_buf);
+ av_free(s->adm_temp_lo);
+ av_free(s->adm_temp_hi);
+ av_free(s->prev_blur_data);
+ av_free(s->blur_data);
+ av_free(s->temp_data);
+ av_free(s->vif_data_buf);
+ av_free(s->vif_temp);
+ }
+
+ ff_dualinput_uninit(&s->dinput);
+}
+
+static const AVFilterPad vmaf_inputs[] = {
+ {
+ .name = "main",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame,
+ },{
+ .name = "reference",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame,
+ .config_props = config_input_ref,
+ },
+ { NULL }
+};
+
+static const AVFilterPad vmaf_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .config_props = config_output,
+ .request_frame = request_frame,
+ },
+ { NULL }
+};
+
+AVFilter ff_vf_vmaf = {
+ .name = "vmaf",
+ .description = NULL_IF_CONFIG_SMALL("Calculate the VMAF between two
video streams."),
+ .init = init,
+ .uninit = uninit,
+ .query_formats = query_formats,
+ .priv_size = sizeof(VMAFContext),
+ .priv_class = &vmaf_class,
+ .inputs = vmaf_inputs,
+ .outputs = vmaf_outputs,
+};
diff --git a/libavfilter/vmaf.h b/libavfilter/vmaf.h
new file mode 100644
index 0000000..6d16f60
--- /dev/null
+++ b/libavfilter/vmaf.h
@@ -0,0 +1,138 @@
+/*
+ * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com>
+ * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Calculate the VMAF between two input videos.
+ */
+
+/** Normalization type */
+const char *norm_type = "linear_rescale";
+
+/** cliping to be applied on vmaf score */
+const double score_clip[2] = {
+ 0.0,
+ 100.0
+};
+
+/** feature vector */
+const char *feature_names[6] = {
+ "VMAF_feature_adm2_score",
+ "VMAF_feature_motion2_score",
+ "VMAF_feature_vif_scale0_score",
+ "VMAF_feature_vif_scale1_score",
+ "VMAF_feature_vif_scale2_score",
+ "VMAF_feature_vif_scale3_score"
+};
+
+const double intercepts[7] = {
+ -0.3092981927591963,
+ -1.7993968597186747,
+ -0.003017198086831897,
+ -0.1728125095425364,
+ -0.5294309090081222,
+ -0.7577185792093722,
+ -1.083428597549764
+};
+
+const double slopes[7] = {
+ 0.012020766332648465,
+ 2.8098077502505414,
+ 0.06264407466686016,
+ 1.222763456258933,
+ 1.5360318811084146,
+ 1.7620864995501058,
+ 2.08656468286432
+};
+
+/** transform constants */
+const double score_transform[3] = {
+ 1.70674692,
+ 1.72643844,
+ -0.00705305
+};
+
+typedef struct {
+ int index;
+ double value;
+} svm_node;
+
+typedef struct {
+ int l;
+ double *y;
+ svm_node **x;
+} svm_problem;
+
+typedef struct {
+ int svm_type;
+ int kernel_type;
+ int degree; /** for poly */
+ double gamma; /** for poly/rbf/sigmoid */
+ double coef0; /** for poly/sigmoid */
+
+ /** these are for training only */
+ double cache_size; /** in MB */
+ double eps; /** stopping criteria */
+ double C; /** for C_SVC, EPSILON_SVR and NU_SVR */
+ int nr_weight; /** for C_SVC */
+ int *weight_label; /** for C_SVC */
+ double* weight; /** for C_SVC */
+ double nu; /** for NU_SVC, ONE_CLASS, and NU_SVR */
+ double p; /** for EPSILON_SVR */
+ int shrinking; /** use the shrinking heuristics */
+ int probability; /** do probability estimates */
+} svm_parameter;
+
+/**
+ * svm_model
+ */
+typedef struct {
+ svm_parameter param; /** parameter */
+ int nr_class; /** number of classes, = 2 in regression/one class
svm */
+ int l; /** total #SV */
+ svm_node **SV; /** SVs (SV[l]) */
+ double **sv_coef; /** coefficients for SVs in decision functions
(sv_coef[k-1][l]) */
+ double *rho; /** constants in decision functions (rho[k*(k-1)/2]) */
+ double *probA; /** pariwise probability information */
+ double *probB;
+ int *sv_indices; /** sv_indices[0,...,nSV-1] are values in
[1,...,num_traning_data] to indicate SVs in the training set */
+
+ /** for classification only */
+
+ int *label; /** label of each class (label[k]) */
+ int *nSV; /** number of SVs for each class (nSV[k]) */
+ /** nSV[0] + nSV[1] + ... + nSV[k-1] = l */
+ int free_sv; /** 1 if svm_model is created by svm_load_model*/
+ /** 0 if svm_model is created by svm_train */
+} svm_model;
+
+
+typedef struct {
+ const svm_node **x;
+ double *x_square;
+
+ // svm_parameter
+ const int kernel_type;
+ const int degree;
+ const double gamma;
+ const double coef0;
+} Kernel;
+
--
2.7.4
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
