Chroma from Luma Intra Prediction for NETVC

draft-egge-netvc-cfl-01

1

Nathan E. Egge

Luc N. Trudeau

David Barr

IETF 100 - Singapore

2017 Nov 15

Mozilla and the Xiph.Org Foundation

What is Chroma from Luma?

Intra prediction tool� No dependencies on other frames

Only available to chroma planes

Predicts chroma using coincident-reconstructed luma pixels

2

What’s New in -01?

​

3

Based on the chroma from luma proposal for AV1

Instead of Daala implementation

​

No longer relies on PVQ

Prediction is done in the spatial domain

Considers only AC contribution of reconstructed luma pixels

Spatial domain equivalent of shape prediction

​

Uses existing chroma DC prediction for DC contribution

Available in AV1, requires no signaling and is more precise

What’s different?

4

[1] draft-midtskogen-netvc-chromapred-02

[2] draft-egge-netvc-cfl-00

How Does it Work?

5

Why use Chroma DC_PRED?

𝛽 is the average chroma reference pixels for a block

​

​

​

​

​

DC_PRED predicts the average value of a block

By computing the average of the neighboring pixels adjacent to the above and left borders of the block

​

No Signaling required

6

What are Scaling Factors (α_Cb, α_Cr)?

Scaling factors set the tone

Scaling factors are in Q3 and range from -2 to 2

Scaling factors are chosen by a rate-constraint search� α = argmin (D(CfL(a)) + 𝜆R(a))� a in A

Scaling factors are signaled to the decoder

​

​

​

7

α_Cb

-α_Cb

α_Cr

-α_Cr

How are Scaling Factors Signaled?

​

A sign can either be [0, -, +]

Signs are jointly coded� using an 8-value¹ CDF

​

​

8

Each non-zero scaling factor is coded � using a 16-value CDF (0,2]� Joint sign used as context

^1. (0,0) is not a valid code as it is equivalent to DC_PRED

UV Mode Selection Example (https://goo.gl/6tKaB8)

9

CFL_PRED 17%

DC_PRED 44.36%

SMOOTH_PRED 4.85%

TM_PRED 7.98%

Ohashi0806shield.y4m

QP = 55

Results (AWCY High Latency)

​

Subset1

​

Objective-1 fast

10

Ref: https://arewecompressedyet.com/?job=master%402017-07-26T10%3A40%3A11.180Z&job=cfl-baseline%402017-07-29T00%3A04%3A47.130Z

​

​

^1. CIEDE2000 is the only metric that combines luma and chroma plane (The distance measured is more perceptually uniform)

Ref: https://arewecompressedyet.com/?job=master%402017-09-13&job=cfl-inter%402017-09-13T14%3A13%3A13.918Z

​

Awesome for Gaming (Twitch dataset)

11

Ref: https://arewecompressedyet.com/?job=no-cfl-twitch-cpu2-60frames%402017-09-18T15%3A39%3A17.543Z&job=cfl-inter-twitch-cpu2-60frames%402017-09-18T15%3A40%3A24.181Z

Notable Mentions