JVET-AL0008 JVET AHG report: Optimization of encoders and receiving systems for machine analysis of coded video content (AHG8) [S. Liu, J. Ström, S. Wang, M. Zhou (AHG chairs)]
Software and Common Test Conditions
AHG 8 related software and documents can be accessed at https://vcgit.hhi.fraunhofer.de/jvet-ahg-ofm. This repository contains two projects: common test conditions, reporting templates with anchor results, evaluation scripts and task networks are available in https://vcgit.hhi.fraunhofer.de/jvet-ahg-ofm/ofm-ctc, and software implementation examples are hosted in https://vcgit.hhi.fraunhofer.de/jvet-ahg-ofm/vtm-ofm.
For this meeting cycle, common test conditions remain unchanged as described in output document JVET-AI2031. Some software implementation examples are available in the repository:
Branch name | Description |
Region-of-Interest-based adaptive QP | |
Foreground/background separation | |
Temporal QP offsets | |
Spatial resampling (Experimental) | |
Post-processing filter (Removed*) | |
Spatial resampling | |
Combined pre- and post-processing | |
Lightweight post-processing | |
Combined software JVET-AB0275, JVET-AC0086, JVET-AJ0178 | |
Pre-analysis based temporal resampling | |
Combined software of adaptive temporal resampling, pre-processing, post-processing and ROI-based adaptive QP |
* Note that the current description of the post-processing algorithm in CDTR A.3 is based on the lightweight design proposed in JVET-AJ0178. Consequently, the reference code for the previous post-processing algorithm, JVET-AG0212, has been removed from the software repository.
It was decided in the last meeting to upload the dense QP bitstreams corresponding to dense QP experiments and results reported in JVET-AK0122 to MPEG content server. They can now be accessed at https://content.mpeg.expert/data/MPEG-05/AHG8/bitstreams_for_JVET-AK0122/.
Technical Report
The eight draft of the technical report (TR) JVET-AK2030 “Optimization of encoders and receiving systems for machine analysis of coded video content (draft 8)” was produced, including the following additions compared with draft 7:
- Text on Packed regions info SEI (to clause 10.5) JVET-AK0141
- Tool combination examples (to annex B) JVET-AK0122
- Tool combination examples (to annex B) JVET-AK0094
The following combined tool examples have been included in the draft TR, besides single tool implementation examples.
Technology 1 | Technology 2 | Technology 3 | Technology 4 |
Adaptive QP (Clause 8.1 and Annex A.1) | Temporal layer QP offset (Clause 8.2) | ||
De-noising filter (Clause 7.5) | Temporal layer QP offset (Clause 8.2) | ||
Pre-processing (Clause 7.2 and Annex A.2) | NNPF (Clause 9.3 and Annex A.3) | ||
4:4:4 coding (Clause 8) | Reduced resolution (Clauses 7.4 and 9.2) | ||
Pre-processing (Clause 7.2 and Annex A.2) | NNPF (Clause 9.3 and Annex A.3) | Adaptive QP (Clause 8.1 and Annex A.1) | |
NNPF (Clause 9.3) | Adaptive QP (Clause 8.1 and Annex A.1) | ||
NNPF (Clause 9.3) | Reduced resolution (Clauses 7.4 and 9.2) | ||
Pre-processing (Clause 7.2 and Annex A.2) | NNPF (Clause 9.3 and Annex A.3) | Adaptive QP (Clause 8.1 and Annex A.1) | QP offset adjustment for higher temporal layers (Clause 8.1) |
Pre-processing (Clause 7.2 and Annex A.2) | NNPF (Clause 9.3 and Annex A.3) | Adaptive QP (Clause 8.1 and Annex A.1) | Temporal resampling (Clause 9.1 and Annex A.4) |
Input contributions
There were 2 input contributions related to AHG 8 mandates (by the time this report was uploaded). They are listed below.
JVET AHG report: Optimization of encoders and receiving systems for machine analysis of coded video content (AHG8) | S. Liu, J. Ström, S. Wang, M. Zhou (AHG chairs) | |
Proposals | ||
AHG8: Dense QP coding results of combining adaptive temporal resampling, pre-processing, ROI-based adaptive QP, QP offset adjustment for higher temporal layers and post-processing algorithms for machine vision | S. Wang, J. Chen, Y. Ye, B. Li (Alibaba), S. Wang (CityUHK) | |
Crosschecks | ||
Recommendations
The AHG recommended to:
- Review all input contributions.
- Continue improving draft TR based on CDTR feedback and other inputs.
- Discuss and plan for finalization of TR (version 1).
- Continue investigating non-normative technologies and their uses for machine vision applications and machine consumptions.
It was agreed that software of the non-normative tools should be attached to the TR (as also requested in context of ballot)
It was clarified that the bitstreams described in JVET-AK0122 and JVET-AL0152 are not intended to be used by JVET (as they are going beyond the CTC document JVET-AI2031), but are delivered as a service to WG 4 such that they can exercise non-normative tools in their own CTC.