AOM AV1

AV1 introduces many new tools, in the following paragraph, the partition modes and the intra prediction tools are presented. Full discription of the coding tools can be found in [1] and [2].

Partition Modes

AV1 allows 10 different partiton modes. Figure below shows the hierarchy of partition modes of different block sizes in AV1. Notice that block Encoded in smaller block size is possible only when split mode is chosen.

img/Partitionhierarchy.png

Partition Hierarchy of AV1

More partition modes help encoder adapt to different pattern and achieve better compression efficiency.

Intra Prediction Tools

AV1 introduce 5 new non-directional modes and up to 56 directions for direction modes depends on the block size.

Directional Mode

AV1 offers 8 main directions ( 0, 45, 67, 90, 113, 135, 157, 203) for directional prediction and each direction can have 6 fine-tuned angles with interval of 3 degree. For example, mode with 45 degree can be fine tuned to 36, 39, 42, 48, 51, 54. In total, there are 56 directions available in compared to 33 directions in HEVC.

For directions need top-right reference pixels, AV1 uses 1 bit of a 8-bit integer to respresent the availabity and stores it in tables for all different subblocks sizes in a 128x128 super block. For example, all the 4x4 blocks in a 128x128 SB needs 1024 bits, which requires 128 8-bit integers in a table. For bottem-left pixels, same method is applied.

_images/intramode.png

PAETH Mode

Paeth mode uses above, left and upper left pixels as reference, also shown in Fig.ref{fig:intra mode}. The actual formula is as follows:

\[base= L+T-TL, P=argmin|x-base|, x \{T,L,TL\}\]

SMOOTH Mode

Smooth mode includes three modes (horizontal, vertical, bi-directional). The prediction formula is as follows:

\[ \begin{align}\begin{aligned}PSMOOTH\_H=w(x)L+(1-w(x))TR\\PSMOOTH\_V=w(y)T+(1-w(y))BL\\PSMOOTH=1/2(PSMOOTH\_H+PSMOOTH\_V)\end{aligned}\end{align} \]

Chroma from Luma(Cfl)

Chroma from Luma uses luma prediction block for chroma prediction. The predicted luma block is first subsampled to the size of chroma then is subtracted by their average and obtain the AC contribution. the AC contribution is multiplied by the scaling factors and added to the DC prediction of chroma. Detailed explanation can be referred to [3].

Palette mode

This tool is especially for screen content videos. It can choose up to 8 base colors for prediction. Only the colors and the index map need to be signaled into the bitstream.

Recursive Intra Filter

This mode divides the block into 4x2 pixels blocks and apply eight 7-tap filters for each pixels in the block.

Intra Block Copy (IntraBC)

This is another mode suitable for screen contents. It works similar to inter prediction but reference to the block in the same frame.

AV1 Performance Review

Many research groups and companies have already tested and evaluated the performance of AV1. Grois, D et al.have compared the performance between AV1 and HEVC. Their result shows HM has better performance than AV1 [4]. On the other hand, the research of MSU Graphics & Media Lab in 2018 has also showed that AV1 outperformed other codecs [5]. Some research also shows the performances between AV1 and HEVC have very similar [6]. BBC compared the efficiency between HEVC HM, AV1, VVC and found AV1 is slightly better than HM [7].

Google also has showed the compression improvement when enabling the experimental tools. Nevertheless, it also shows the encoding speed is extremely slow. Sethuraman et al. cite [8] also shows AV1 has better performance than other video codecs but has very slow encoding time. Similar results are also performed by Facebook [9] and Bitmovin [10].

[1]Urvang Joshi, Debargha Mukherjee, Jingning Han, Yue Chen, Sarah Parker, Hui Su, Angie Chiang, Yaowu Xu, Zoe Liu, Yunqing Wang, Jim Bankoski, Chen Wang, Emil Keyder, “Novel inter and intra prediction tools under consideration for the emerging AV1 video codec,” Proc. SPIE 10396, Applications of Digital Image Processing XL, 103960F (19 September 2017);
[2]
  1. Chen et al., “An Overview of Core Coding Tools in the AV1 Video Codec,” 2018 Picture Coding Symposium (PCS), San Francisco, CA, 2018, pp. 41-45.
[3]
  1. Trudeau, N. Egge and D. Barr, “Predicting Chroma from Luma in AV1,” 2018 Data Compression Conference, Snowbird, UT, 2018, pp. 374-382.
[4]Dan Grois, Tung Nguyen, and Detlev Marpe “Performance comparison of AV1, JEM, VP9, and HEVC encoders,” Proc. SPIE 10396, Applications of Digital Image Processing XL, 103960L (8 February 2018);
[5]https://www.compression.ru/video/codec_comparison/hevc_2018/
[6]
  1. Akyazi and T. Ebrahimi, “Comparison of Compression Efficiency between HEVC/H.265, VP9 and AV1 based on Subjective Quality Assessments,” 2018 Tenth International Conference on Quality of Multimedia Experience (QoMEX), Cagliari, 2018, pp. 1-6.
[7]https://www.bbc.co.uk/rd/blog/2019-05-av1-codec-streaming-processing-hevc-vvc
[8]
  1. Sethuraman, C. Rajan and K. Patankar, “Analysis of the Emerging AOMedia AV1 Video Coding Format for OTT Use-Cases,” in SMPTE Motion Imaging Journal, vol. 127, no. 5, pp. 44-50, June 2018.
[9]https://engineering.fb.com/video-engineering/av1-beats-x264-and-libvpx-vp9-in-practical-use-case/
[10]https://bitmovin.com/av1-multi-codec-dash-dataset/