Back to Search Document details
41st Meeting: by teleconference, CH, January 2026 2026-01-20 18:10
[Non-EE2] Intra lfnstIdx Prediction
Abstract
In this contribution, a new way to signal the intra LFNST index is proposed. The proposed prediction method tests the possible inverse transforms and utilizes the boundary continuity metric as the prediction criterion. The proposed method is tested on ECM-19.0, and the simulation results are summarized as follows:
JVET-AO0218 Intra lfnstIdx Prediction [Y. Wang, R. Chernyak, S. Liu (Tencent)] [late]

In this contribution, a new way to signal the intra LFNST index is proposed. The proposed prediction method tests the possible inverse transforms and utilizes the boundary continuity metric as the prediction criterion. The proposed method is tested on ECM-19.0, and the simulation results are summarized as follows:

AI: -0.01%(Y), -0.17%(U), -0.14%(V), 98.4% (EncT), 98.9% (DecT)

A diagram of a cost calculation

AI-generated content may be incorrect.A diagram of a function

AI-generated content may be incorrect.

Invocation of the proposed also has the following conditions:

  1. If MTSS is allowed, the proposed method is disabled.
  2. If the proposed method is allowed, sign prediction is disabled.

The proposal is an additional signalling method because when the above conditions are met, we would need the existing signalling method in ECM 19.0.

Non-CTC results (where sign prediction is disabled) were not available at the time of presentation.

The proponent commented that the proposed method wouldn’t be effective for inter LFNST.

Multiple experts expressed interest in studying in EE given the tradeoff is favourable – that is, although gain is small, there is some reduction in encoding time.

It was agreed to investigate this in an EE, the following aspects should be studied, RA results are to be provided and non-CTC results with sign prediction off are requested.

Decisions
non-CTC results with sign prediction off are requested.
Citation