Back to Search Document details
30th Meeting: Antalya, TR, April 2023 2023-04-23 12:45
AhG12: Shifting Quantizer Center
Abstract
This contribution proposes a straightforward implementation of recently proposed Latent-shift algorithm on the end-to-end compression [1] to the conventional compression. This technique proposes to add some offset value which is driven by gradient of the entropy on the reconstructed transform coefficient.
JVET-AD0251 AhG12: Shifting Quantizer Center [M. Balcilar, K. Naser, F. Galpin, F. Le Léannec (InterDigital)] [late]

This contribution proposes an implementation of a recently proposed latent-shift algorithm on the end-to-end compression to the conventional compression. This technique proposes to add some offset value which is driven by gradient of the entropy on the reconstructed transform coefficients.

The following results are obtained:

  • AI: {-0.09%, -0.12%, -0.13%; 99%, 100%}.
  • RA: {%, %, %; %, %}.

Gain is small, but several experts expressed interest, as it is extremely simple to implement (no change in quantization, “post processing” of dequantized output), and the gain is consistent over classes.

It was asked if there is dependency on bit depth.

It was agreed to investigate this in an EE. It was suggested to consider checking tendency of performance in other QP ranges.

Decisions
It was agreed to investigate this in an EE. It was suggested to consider checking tendency of performance in other QP ranges.
Citation