Add codec "avifenc" with various speed param values. Using "avifenc" …

…binary -- a libavif-based tool.

.gitignore

@@ -27,4 +27,6 @@ parallel_*txt
 4*.png
 __pycache__/
 stats.json
-stats.log
+stats.log
+bdrates_*.db.txt
+bitrate_savings_*.db.txt

Dockerfile

@@ -27,7 +27,7 @@ ENV FFMPEG_VERSION=4.2.1 \
  OGG_VERSION=1.3.2 \
  OPUS_VERSION=1.2 \
  OPENJPEG_VERSION=2.3.1 \
- THEORA_VERSION=1.1.1 \
+# THEORA_VERSION=1.1.1 \
  VORBIS_VERSION=1.3.5 \
  VPX_VERSION=1.8.1 \
  WEBP_VERSION=1.0.3 \
@@ -46,7 +46,7 @@ ENV FFMPEG_VERSION=4.2.1 \
 ARG OGG_SHA256SUM="e19ee34711d7af328cb26287f4137e70630e7261b17cbe3cd41011d73a654692 libogg-1.3.2.tar.gz"
 ARG OPUS_SHA256SUM="77db45a87b51578fbc49555ef1b10926179861d854eb2613207dc79d9ec0a9a9 opus-1.2.tar.gz"
 ARG VORBIS_SHA256SUM="6efbcecdd3e5dfbf090341b485da9d176eb250d893e3eb378c428a2db38301ce libvorbis-1.3.5.tar.gz"
-ARG THEORA_SHA256SUM="40952956c47811928d1e7922cda3bc1f427eb75680c3c37249c91e949054916b libtheora-1.1.1.tar.gz"
+# ARG THEORA_SHA256SUM="40952956c47811928d1e7922cda3bc1f427eb75680c3c37249c91e949054916b libtheora-1.1.1.tar.gz"
 ARG XVID_SHA256SUM="4e9fd62728885855bc5007fe1be58df42e5e274497591fec37249e1052ae316f xvidcore-1.3.4.tar.gz"
 ARG FREETYPE_SHA256SUM="5d03dd76c2171a7601e9ce10551d52d4471cf92cd205948e60289251daddffa8 freetype-2.5.5.tar.gz"
 ARG LIBVIDSTAB_SHA256SUM="14d2a053e56edad4f397be0cb3ef8eb1ec3150404ce99a426c4eb641861dc0bb v1.1.0.tar.gz"
@@ -66,15 +66,18 @@ RUN buildDeps="autoconf \
  gperf \
  imagemagick \
  libexpat1-dev \
+ libpng-dev \
+ libjpeg-dev \
  libssl-dev \
  libtool \
  make \
- nasm \
+ ninja-build \
  patchelf \
  perl \
  pkg-config \
  python3 \
  python3-pip \
+ re2c \
  subversion \
  unzip \
  wget \
@@ -87,6 +90,33 @@ RUN buildDeps="autoconf \
 # SQLite3 is to store metrics.
 RUN apt-get install -y sqlite3 libsqlite3-dev
 
+# Install nasm 2.14
+RUN curl -L https://download.videolan.org/contrib/nasm/nasm-2.14.tar.gz | tar xvz && \
+ cd nasm-2.14 && \
+ ./configure && make -j2 && make install && \
+ cd .. && \
+ nasm --version
+
+# Install latest ninja
+RUN git clone -b release https://github.com/ninja-build/ninja.git && \
+ cd ninja/ && \
+ python3 ./configure.py && \
+ ninja && \
+ cp ninja `which ninja` && \
+ cd .. && \
+ ninja --version
+
+# AVIFENC
+RUN mkdir -p /tools && \
+ cd /tools && \
+ git clone https://github.com/AOMediaCodec/libavif && \
+ cd /tools/libavif/ext && \
+ $SHELL ./aom.cmd && \
+ cd /tools/libavif && \
+ mkdir build && cd build && \
+ cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=OFF -DAVIF_CODEC_AOM=ON -DAVIF_LOCAL_AOM=ON -DAVIF_CODEC_DAV1D=OFF -DAVIF_LOCAL_DAV1D=OFF -DAVIF_CODEC_RAV1E=OFF -DAVIF_LOCAL_RAV1E=OFF -DAVIF_CODEC_LIBGAV1=OFF -DAVIF_LOCAL_LIBGAV1=OFF -DAVIF_BUILD_TESTS=1 -DAVIF_BUILD_APPS=1 && \
+ ninja
+
 ## opencore-amr https://sourceforge.net/projects/opencore-amr/
 # RUN \
 # DIR=/tmp/opencore-amr && \
@@ -166,17 +196,17 @@ RUN \
  rm -rf ${DIR}
 
 ### libtheora http:https://www.theora.org/
-RUN \
- DIR=/tmp/theora && \
- mkdir -p ${DIR} && \
- cd ${DIR} && \
- curl -sLO http:https://downloads.xiph.org/releases/theora/libtheora-${THEORA_VERSION}.tar.gz && \
- echo ${THEORA_SHA256SUM} | sha256sum --check && \
- tar -zx --strip-components=1 -f libtheora-${THEORA_VERSION}.tar.gz && \
- ./configure --prefix="${PREFIX}" --with-ogg="${PREFIX}" --enable-shared && \
- make && \
- make install && \
- rm -rf ${DIR}
+#RUN \
+# DIR=/tmp/theora && \
+# mkdir -p ${DIR} && \
+# cd ${DIR} && \
+# curl -sLO http:https://downloads.xiph.org/releases/theora/libtheora-${THEORA_VERSION}.tar.gz && \
+# echo ${THEORA_SHA256SUM} | sha256sum --check && \
+# tar -zx --strip-components=1 -f libtheora-${THEORA_VERSION}.tar.gz && \
+# ./configure --prefix="${PREFIX}" --with-ogg="${PREFIX}" --enable-shared && \
+# make && \
+# make install && \
+# rm -rf ${DIR}
 
 ### libvpx https://www.webmproject.org/code/
 RUN \
@@ -387,7 +417,7 @@ RUN \
 # --enable-libmp3lame \
  --enable-libopenjpeg \
  --enable-libopus \
- --enable-libtheora \
+# --enable-libtheora \
  --enable-libvorbis \
  --enable-libvpx \
 # --enable-libwebp \

analyze_encoding_results.py

@@ -98,8 +98,9 @@ def main(argv):
  baseline_codec = 'jpeg'
  sub_sampling_arr = ['420', '444']
  codecs = ['jpeg-mse', 'jpeg-ms-ssim', 'jpeg-im', 'jpeg-hvs-psnr', 'webp', 'kakadu-mse', 'kakadu-visual', 'openjpeg',
- 'hevc', 'avif-mse', 'avif-ssim']
- plot_list = ['webp', 'kakadu-mse', 'kakadu-visual', 'hevc', 'avif-mse', 'avif-ssim']
+ 'hevc', 'avif-mse', 'avif-ssim', 'avifenc-sp-0', 'avifenc-sp-2', 'avifenc-sp-4', 'avifenc-sp-8']
+ # plot_list = ['webp', 'kakadu-mse', 'kakadu-visual', 'hevc', 'avif-mse', 'avif-ssim']
+ plot_list = ['webp', 'avif-mse', 'avifenc-sp-0', 'avifenc-sp-2', 'avifenc-sp-4', 'avifenc-sp-8']
  color_list = ['tab:blue', 'tab:orange', 'tab:green', 'tab:red', 'tab:purple', 'tab:brown', 'tab:pink', 'tab:gray', 'tab:olive', 'tab:cyan']
  marker_list = ['o', 'v', '>', '<', 's', 'p', 'd', '4', 'P', 'X']
  assert len(color_list) == len(marker_list) # 10 curves on one plot is the limit, beyond that is sensory overload

compute_BD_rates.py

@@ -112,8 +112,8 @@ def main(argv):
  baseline_codec = 'jpeg'
  sub_sampling_arr = ['420', '444']
  codecs = ['jpeg-mse', 'jpeg-ms-ssim', 'jpeg-im', 'jpeg-hvs-psnr', 'webp', 'kakadu-mse', 'kakadu-visual', 'openjpeg',
- 'hevc', 'avif-mse', 'avif-ssim']
- metrics_for_BDRate = ['vmaf', 'ssim', 'ms_ssim', 'vif', 'psnr_y', 'psnr_avg']
+ 'hevc', 'avif-mse', 'avif-ssim', 'avifenc-sp-0', 'avifenc-sp-2', 'avifenc-sp-4', 'avifenc-sp-8']
+ metrics_for_BDRate = ['vmaf', 'ssim', 'ms_ssim', 'vif', 'adm2', 'psnr_y', 'psnr_avg']
  for sub_sampling in sub_sampling_arr:
  bdrates_various_metrics = dict()
  black_list_source_various_metrics = dict()

script_compress_parallel.py

@@ -87,6 +87,8 @@
  CodecType('hevc', True, 10, 51, 1, '444'),
  CodecType('hevc', True, 10, 51, 1, '444u'),
 
+ # The codecs 'avif-mse' and 'avif-ssim' write out a "single-frame video" file in IVF format,
+ # so technically not a proper AVIF. But they accomplish "full-range" YUV encoding similar to a "proper" AVIF.
  CodecType('avif-mse', True, 8, 63, 1, '420'),
  CodecType('avif-mse', True, 8, 63, 1, '444'),
  CodecType('avif-mse', True, 8, 63, 1, '444u'),
@@ -95,6 +97,24 @@
  CodecType('avif-ssim', True, 8, 63, 1, '444'),
  CodecType('avif-ssim', True, 8, 63, 1, '444u'),
 
+ # Codecs starting with 'avifenc' prefix write out a proper AVIF; avifenc is a libavif-based tool.
+ # https://github.com/AOMediaCodec/libavif
+ CodecType('avifenc-sp-0', True, 8, 62, 1, '420'),
+ CodecType('avifenc-sp-0', True, 8, 62, 1, '444'),
+ CodecType('avifenc-sp-0', True, 8, 62, 1, '444u'),
+
+ CodecType('avifenc-sp-2', True, 8, 62, 1, '420'),
+ CodecType('avifenc-sp-2', True, 8, 62, 1, '444'),
+ CodecType('avifenc-sp-2', True, 8, 62, 1, '444u'),
+
+ CodecType('avifenc-sp-4', True, 8, 62, 1, '420'),
+ CodecType('avifenc-sp-4', True, 8, 62, 1, '444'),
+ CodecType('avifenc-sp-4', True, 8, 62, 1, '444u'),
+
+ CodecType('avifenc-sp-8', True, 8, 62, 1, '420'),
+ CodecType('avifenc-sp-8', True, 8, 62, 1, '444'),
+ CodecType('avifenc-sp-8', True, 8, 62, 1, '444u'),
+
 )
 
 LOGGER = logging.getLogger('image.compression')
@@ -601,6 +621,38 @@ def f(param, codec, image, width, height, temp_folder, subsampling):
  cmd = ['aomdec', '--rawvideo', '-o', decoded_yuv, encoded_file]
  my_exec(cmd)
 
+ elif codec in ['avifenc-sp-0', 'avifenc-sp-2', 'avifenc-sp-4', 'avifenc-sp-8'] and subsampling in ['420', '444u', '444']:
+ min_QP = int(param) - 1
+ max_QP = int(param) + 1
+ info = codec.split('-')
+ speed = info[2]
+
+ cmd = ['ffmpeg', '-y', '-i', image, '-pix_fmt', get_pixel_format_for_encoding(subsampling), source_yuv]
+ my_exec(cmd)
+
+ source_y4m = get_filename_with_temp_folder(temp_folder, 'source.y4m')
+ cmd = ['ffmpeg', '-y', '-f', 'rawvideo', '-pix_fmt', get_pixel_format_for_encoding(subsampling),
+ '-s:v', '%s,%s' % (width, height), '-i', source_yuv, source_y4m]
+ my_exec(cmd)
+
+ encoded_file = get_filename_with_temp_folder(temp_folder, 'temp.avif')
+ # bit-depth and yuv subsampling are maintained for y4m input
+ cmd = ['/tools/libavif/build/avifenc', source_y4m, encoded_file, '--nclx', '1/13/1', '--min', str(min_QP),
+ '--max', str(max_QP), '--speed', speed, '--codec', 'aom', '--jobs', '4']
+ my_exec(cmd)
+
+ decoded_y4m = get_filename_with_temp_folder(temp_folder, 'decoded.y4m')
+ cmd = ['/tools/libavif/build/avifdec', '--codec', 'aom', encoded_file, decoded_y4m]
+ my_exec(cmd)
+
+ # explicitly convert to 420 or 444 depending on the case from the y4m
+ cmd = ['ffmpeg', '-y', '-i', decoded_y4m, '-pix_fmt', get_pixel_format_for_encoding(subsampling),
+ decoded_yuv]
+ my_exec(cmd)
+
+ if subsampling == '444u':
+ source_yuv, decoded_yuv = convert_420_to_444_source_and_decoded(source_yuv, decoded_yuv, image, width, height, subsampling)
+
  else:
  raise RuntimeError('Unsupported codec and subsampling ' + codec + ' / ' + subsampling)