Back to Search Document details

JVET-AF0014 3 i (m64948)

32nd Meeting: Hannover, DE, October 2023 2023-10-12 08:17

JVET AHG report: NNVC software development (AHG14)

Authors: Franck Galpin Yun Li Yun Li Jay Shingala Liqiang Wang Zhihuang Xie Jens-Rainer Ohm

Abstract

This report summarizes the activities of the AHG14 on NNVC software development that have taken place between the 31st and 32nd JVET meetings.

Meeting Notes

JVET-AF0014 JVET AHG report: NNVC software development (AHG14) [F. Galpin (chair), Y. Li, Y. Li, J. Shingala, L. Wang, Z. Xie (vice chairs)]

Software development

Location

NNVC repository is located at https://vcgit.hhi.fraunhofer.de/jvet-ahg-nnvc/VVCSoftware_VTM

NNVC software is based on VTM-11.0 with enabled MCTF including the update from JVET-V0056, GOP32, and enabling deblocking in the RDO.

NNVC-6.0 anchor at https://vcgit.hhi.fraunhofer.de/jvet-ahg-nnvc/VVCSoftware_VTM is used for NNVC performance evaluation.

Software changes

Several commits were merged in the NNVC repository.

The following changes were integrated:

HOP related
- JVET-AE0191 HOP models
- JVET-AE0289 HOP stage 3 training process
- JVET-AE0291 HOP stage 3 dataset
- JVET-AE0214 SADL v6
- clarify documentation
- naive quantization script
fixes:
- training hyper-parameters
- 3 issues related to residual scaling in HOP (available in NNVC-6.1)
- avoid log overwrite when resuming training
- port JVET-AE0180 (memory reduction)
- correct resume training hyper-parameters reloading
cleaning:
- remove dead code
- remove dead features

Software versions

NNVC-6.1 was tagged September 25^th, 2023 (fix)
NNVC-6.0 was tagged September 6^th, 2023.
NNVC-5.1 was tagged July 19^th, 2023.
NNVC-5.0 was tagged May 11^th, 2023.
NNVC-3.0 (a.k.a VTM-11.0_nnvc3.0) was tagged December 1^st 2022.
NCS-1.0 (a.k.a NNVC-3.0wip2) was tagged September 4^th 2022 (first release containing the FilterSets, using NNVC 2.0 as a base).
VTM-11.0_nnvc-2.0 was tagged August 4^th 2022 (add deblocking in RDO).
VTM-11.0_nnvc-1.0 was tagged May 6^th 2021 (VTM-11.0 base with MCTF enabled).

CTC performance

See configurations section for the naming convention.

Comparison to VTM

Comparison to previous version – NNVC-5.0 vs NNVC-6.0 VTM mode

NNVC-6.0 in VTM mode performance are the same as the ones in NNVC-5.0 VTM (only changes are memory consumption).

	Random access Main 10
	BD-rate Over nnvc5.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	99%
Class A2	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	99%
Class B	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class C	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class E
Overall	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class D	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class F	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%

	Low delay B Main 10
	BD-rate Over nnvc5.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1
Class A2
Class B	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	99%
Class C	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Class E	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Overall	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Class D	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class F	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	97%	98%

	All Intra Main 10
	BD-rate Over nnvc5.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Class A2	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Class B	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	100%	99%
Class C	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	98%
Class E	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	100%
Overall	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	99%	99%
Class D	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	100%	97%
Class F	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	98%	96%

NNVC-6.0 VTM mode vs NNVC-6.0 anchor

The NNVC-6.0 anchor includes LC filter and Intra Prediction tools activated. Results are the same as for NNVC-5.0 anchor (results are copied here for reference)

	Random access Main 10
	BD-rate Over nnvc4.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-7.23%	-4.86%	-6.00%	-8.72%	-4.05%	-5.06%	134%	7667%
Class A2	-6.56%	-5.78%	-5.09%	-6.53%	-4.14%	-2.90%	130%	7183%
Class B	-6.29%	-8.19%	-7.53%	-5.97%	-6.06%	-5.18%	132%	7834%
Class C	-6.62%	-10.53%	-9.55%	-6.26%	-7.91%	-5.58%	132%	7422%
Class E
Overall	-6.62%	-7.66%	-7.28%	-6.71%	-5.77%	-4.81%	132%	7557%
Class D	-7.57%	-9.45%	-9.68%	-5.76%	-6.81%	-5.94%	129%	8193%
Class F	-3.07%	-5.88%	-4.94%	-3.05%	-4.76%	-3.54%	148%	9739%

	Low delay B Main 10
	BD-rate Over nnvc4.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1
Class A2
Class B	-4.85%	-8.13%	-7.84%	-5.15%	-6.16%	-6.65%	134%	6812%
Class C	-5.24%	-11.13%	-9.00%	-5.61%	-10.74%	-4.02%	126%	6233%
Class E	-4.88%	-4.09%	-5.17%	-5.64%	-2.18%	-3.36%	162%	9877%
Overall	-4.98%	-8.12%	-7.56%	-5.42%	-6.69%	-4.95%	138%	7257%
Class D	-6.54%	-10.76%	-9.77%	-5.75%	-9.48%	-6.83%	125%	6400%
Class F	-2.05%	-4.88%	-4.46%	-2.29%	-3.08%	-4.12%	148%	8941%

	All Intra Main 10
	BD-rate Over nnvc4.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-8.46%	-7.23%	-8.67%	-10.00%	-8.00%	-7.50%	210%	5134%
Class A2	-6.61%	-8.04%	-7.94%	-6.90%	-6.44%	-5.44%	204%	4416%
Class B	-7.03%	-8.58%	-8.71%	-7.04%	-7.61%	-6.99%	203%	4493%
Class C	-7.34%	-9.76%	-9.92%	-7.61%	-8.38%	-7.64%	190%	3770%
Class E	-10.29%	-9.04%	-8.85%	-10.14%	-7.52%	-6.38%	199%	5149%
Overall	-7.81%	-8.60%	-8.87%	-8.16%	-7.64%	-6.86%	201%	4507%
Class D	-7.57%	-8.11%	-9.33%	-7.16%	-6.81%	-7.56%	180%	3760%
Class F	-4.55%	-5.97%	-5.65%	-4.24%	-4.75%	-4.38%	155%	3961%

NNVC-6.0 VTM mode vs NNVC-6.0 VTM mode + HOP (HOP only)

	Random access Main 10
	BD-rate Over nnvc6.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-10.52%	-20.62%	-25.85%	-11.91%	-22.85%	-27.10%	176%	49507%
Class A2	-10.79%	-23.75%	-22.70%	-10.55%	-23.08%	-19.47%	169%	47472%
Class B	-9.65%	-25.88%	-24.33%	-9.17%	-25.28%	-24.84%	171%	49184%
Class C	-10.49%	-27.44%	-27.39%	-9.82%	-24.33%	-25.44%	150%	44716%
Class E
Overall	-10.27%	-24.82%	-25.12%	-10.17%	-24.10%	-24.38%	166%	47674%
Class D	-11.88%	-26.75%	-27.45%	-9.11%	-24.06%	-23.76%	145%	41584%
Class F	-5.81%	-18.45%	-17.28%	-6.60%	-19.43%	-18.41%	218%	18932%

	Low delay B Main 10
	BD-rate Over nnvc6.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1
Class A2
Class B	-9.46%	-20.43%	-14.83%	-9.63%	-21.13%	-17.32%	166%	48578%
Class C	-10.61%	-25.51%	-22.61%	-10.89%	-25.00%	-21.30%	148%	44637%
Class E	-11.79%	-15.78%	-6.88%	-12.19%	-16.97%	-7.79%	266%	54934%
Overall	-10.43%	-20.96%	-15.43%	-10.69%	-21.38%	-16.27%	180%	48701%
Class D	-11.81%	-23.44%	-16.11%	-10.59%	-23.09%	-14.53%	141%	41240%
Class F	-6.98%	-19.22%	-14.15%	-8.44%	-20.83%	-20.68%	219%	23962%

	All Intra Main 10
	BD-rate Over nnvc6.0_VTM-anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-7.20%	-18.27%	-22.13%	-8.54%	-20.42%	-23.39%	180%	34113%
Class A2	-7.09%	-20.09%	-18.85%	-7.96%	-19.89%	-16.46%	142%	28458%
Class B	-6.98%	-19.22%	-19.78%	-7.01%	-20.82%	-21.74%	134%	26660%
Class C	-8.08%	-18.41%	-21.09%	-7.96%	-20.09%	-22.71%	118%	18245%
Class E	-10.45%	-20.65%	-21.76%	-10.85%	-20.11%	-23.27%	140%	29315%
Overall	-7.86%	-19.27%	-20.64%	-8.27%	-20.32%	-21.60%	139%	26224%
Class D	-7.92%	-17.77%	-21.44%	-7.11%	-21.03%	-22.49%	114%	16314%
Class F	-5.27%	-15.78%	-15.91%	-5.35%	-17.94%	-18.72%	120%	21783%

NNVC-6.0 VTM mode vs NNVC-6.0 + HOP

	Random access Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-12.25%	-21.90%	-26.99%	-13.64%	-23.87%	-28.04%	192%	49788%
Class A2	-11.87%	-24.22%	-23.45%	-11.67%	-23.33%	-20.04%	183%	47404%
Class B	-11.02%	-26.65%	-25.11%	-10.45%	-25.97%	-24.94%	187%	49058%
Class C	-11.78%	-28.42%	-28.35%	-11.11%	-25.13%	-25.96%	166%	44554%
Class E
Overall	-11.64%	-25.69%	-26.02%	-11.51%	-24.80%	-24.85%	181%	47628%
Class D	-12.99%	-27.60%	-28.21%	-10.31%	-24.82%	-24.56%	159%	41349%
Class F	-6.64%	-18.95%	-17.80%	-7.35%	-19.89%	-18.68%	234%	18908%

	Low delay B Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM		V-MSIM	EncT		DecT CPU
Class A1	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!		#VALUE!	#DIV/0!		#DIV/0!
Class A2	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!		#VALUE!	#DIV/0!		#DIV/0!
Class B	-9.87%	-20.94%	-15.27%	-10.13%	-21.91%		-18.18%	174%		48596%
Class C	-10.84%	-25.68%	-22.82%	-11.11%	-25.25%		-21.39%	153%		44563%
Class E	-11.89%	-16.58%	-6.94%	-12.09%	-17.05%		-7.82%	269%		54898%
Overall	-10.70%	-21.43%	-15.70%	-10.94%	-21.81%		-16.66%	186%		48674%
Class D	-11.97%	-23.63%	-15.96%	-10.74%	-22.83%		-13.85%	147%		41272%
Class F	-7.11%	-19.37%	-14.40%	-8.62%	-21.48%		-21.24%	225%		23976%

	All Intra Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM		U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-10.80%	-21.61%	-25.21%	-12.42%		-24.08%	-26.37%	257%	34651%
Class A2	-9.50%	-22.21%	-20.70%	-10.28%		-22.04%	-18.20%	221%	29092%
Class B	-9.67%	-21.93%	-22.37%	-9.68%		-23.46%	-23.99%	211%	27305%
Class C	-10.66%	-21.34%	-23.99%	-10.48%		-22.56%	-25.23%	187%	18934%
Class E	-14.31%	-24.62%	-25.51%	-14.72%		-23.78%	-26.51%	212%	29758%
Overall	-10.82%	-22.24%	-23.45%	-11.25%		-23.18%	-24.12%	214%	26851%
Class D	-10.32%	-20.91%	-24.43%	-9.53%		-23.50%	-25.27%	176%	17077%
Class F	-7.18%	-17.41%	-17.55%	-7.14%		-19.12%	-20.80%	161%	22168%

Comparison to NNVC-6.0 anchor

NNVC-6.0 anchor vs NNVC-6.0 HOP

	Random access Main 10
	BD-rate Over NNVC-6.0 anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-5.41%	-17.63%	-22.03%	-5.35%	-20.16%	-23.98%	145%	679%
Class A2	-5.71%	-19.22%	-19.10%	-5.49%	-19.70%	-17.46%	142%	695%
Class B	-5.06%	-19.76%	-18.79%	-4.75%	-21.02%	-20.82%	141%	640%
Class C	-5.52%	-19.47%	-20.33%	-5.15%	-18.40%	-21.14%	128%	637%
Class E
Overall	-5.38%	-19.15%	-19.91%	-5.12%	-19.88%	-20.86%	138%	658%
Class D	-5.84%	-19.48%	-19.97%	-4.76%	-18.71%	-19.54%	129%	567%
Class F	-3.74%	-13.98%	-13.51%	-4.46%	-16.07%	-15.83%	158%	207%

	Low delay B Main 10
	BD-rate Over NNVC-6.0 anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!
Class A2	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!
Class B	-5.29%	-13.80%	-7.93%	-5.25%	-16.62%	-12.00%	128%	606%
Class C	-5.91%	-16.15%	-15.00%	-5.80%	-16.41%	-18.24%	120%	617%
Class E	-7.28%	-12.89%	-2.29%	-6.77%	-15.05%	-5.00%	163%	434%
Overall	-5.99%	-14.36%	-8.88%	-5.81%	-16.16%	-12.33%	133%	561%
Class D	-5.79%	-14.12%	-6.95%	-5.27%	-14.49%	-7.18%	117%	556%
Class F	-5.19%	-15.13%	-10.53%	-6.60%	-18.51%	-17.78%	149%	209%

	All Intra Main 10
	BD-rate Over NNVC-6.0 anchor
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-2.58%	-15.45%	-18.02%	-2.61%	-17.91%	-20.40%	119%	592%
Class A2	-3.12%	-15.30%	-13.68%	-3.64%	-16.63%	-13.39%	106%	602%
Class B	-2.88%	-14.39%	-14.82%	-2.83%	-17.15%	-18.16%	102%	540%
Class C	-3.63%	-12.62%	-15.24%	-3.12%	-15.61%	-18.81%	96%	464%
Class E	-4.52%	-16.74%	-18.11%	-5.00%	-17.14%	-21.34%	105%	529%
Overall	-3.31%	-14.72%	-15.81%	-3.35%	-16.85%	-18.41%	104%	538%
Class D	-2.98%	-13.49%	-16.04%	-2.54%	-17.60%	-18.62%	94%	393%
Class F	-2.82%	-12.02%	-12.45%	-2.94%	-14.92%	-16.77%	100%	495%

Comparison to NNVC-6.0 HOP

NNVC-6.0 anchor NNVC-6.1 HOP

A bug fix was performed on top of NNVC 6.0 and tagged 6.1. It only affects the HOP results. Please note that timing is not reliable.

	Random access Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-0.08%	-0.57%	-0.40%	0.01%	-0.42%	-0.17%	140%	153%
Class A2	-0.04%	-0.15%	0.07%	-0.08%	-0.08%	0.02%	139%	#NUM!
Class B	-0.05%	0.00%	-0.53%	-0.14%	-0.36%	-0.62%	142%	159%
Class C	0.03%	-0.24%	-0.32%	0.02%	-0.29%	-0.42%	143%	170%
Class E
Overall	-0.03%	-0.21%	-0.33%	-0.06%	-0.30%	-0.35%	141%	#NUM!
Class D	0.08%	-0.33%	-0.45%	0.12%	-0.26%	-0.22%	141%	170%
Class F	-0.16%	-0.71%	-0.50%	0.17%	-0.48%	-0.44%	143%	175%

	Low delay B Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#DIV/0!	#DIV/0!
Class A2	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#VALUE!	#DIV/0!	#DIV/0!
Class B	0.03%	-0.38%	-3.75%	0.13%	-0.96%	-3.75%	138%	#NUM!
Class C	-0.01%	-0.17%	-0.89%	0.03%	-0.11%	-1.18%	139%	166%
Class E	0.06%	-0.82%	-0.70%	-0.07%	-0.61%	-0.61%	145%	150%
Overall	0.03%	-0.42%	-2.04%	0.05%	-0.59%	-2.11%	140%	#NUM!
Class D	-0.06%	-2.23%	-2.57%	0.13%	-3.51%	-2.84%	138%	170%
Class F	-0.07%	-0.06%	-0.74%	0.06%	-0.56%	-1.02%	143%	#NUM!

	All Intra Main 10
	BD-rate Over NNVC-6.0 VTM
	Y-PSNR	U-PSNR	V-PSNR	Y-MSIM	U-MSIM	V-MSIM	EncT	DecT CPU
Class A1	-0.14%	-0.17%	0.75%	-0.16%	0.02%	0.52%	144%	147%
Class A2	-0.02%	-0.13%	-0.04%	-0.04%	0.19%	-0.07%	143%	146%
Class B	-0.06%	0.04%	-0.09%	-0.07%	-0.04%	-0.18%	141%	149%
Class C	-0.09%	-0.92%	-0.90%	-0.12%	-0.82%	-0.86%	140%	163%
Class E	-0.14%	0.78%	0.56%	-0.08%	0.56%	0.37%	142%	152%
Overall	-0.09%	-0.11%	-0.01%	-0.09%	-0.07%	-0.10%	142%	152%
Class D	-0.05%	-0.12%	-0.31%	0.01%	-0.11%	-0.58%	137%	166%
Class F	-0.23%	-0.47%	1.12%	-0.08%	-0.72%	1.86%	131%	165%

Other tools

Other results remain the same as tools were not changed.

Discussions

Previous discussions

The following points were discussed:

Final training strategy for HOP model: final training and models were decided.
Some aspects of encoder optimization (eg RDO models) are not yet aligned with HOP model.

Configurations

The following configurations is used to generate the different NNVC results.

The column “tested” is read as follow:

Y: the configuration has been tested using the new NNVC-6.0 software
P: the results are the ones from previous NNVC software basis
N: not tested.

The column “xcheck” is read as follow:

Y: the test has been cross-checked
P: no cross-checked performed but results are consistent with previous version on NNVC
N: no cross-check available

Name	Tools	Configuration	Tested	Xcheck
NNVC-6.0 VTM mode	none	encoder_xxx_vtm.cfg	Y	Y
NNVC 6.0 Anchor/EE1	Intra Pred + LC filterset	encoder_xxx_nnvc.cfg	Y	Y
NNVC-6.0. HOP	Intra Pred + HOP	encoder_xxx_nnvc.cfg + nn-based/HOP.cfg	Y
HOP only	HOP	encoder_xxx_vtm.cfg + nn-based/HOP.cfg	Y
set0	Loop filter set #0	encoder_xxx_vtm.cfg + nn-based/NnlfOption_1.cfg	P	Y
set1	Loop filter set #1	encoder_xxx_vtm.cfg + nn-based/NnlfOption_2.cfg	P	Y
Intra	Intra prediction	nn-based/encoder_xxx_nnvc.cfg	P	Y
set0+intra	Loop filter set #0+ Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_1.cfg	P	Y
set1+intra	Loop filter set #1+ Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_2.cfg	P	Y
set0+rdo+intra	Loop filter set #0 + Rdo + Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_1.cfg + --EncNnlfOpt=1	P	Y
set1+rdo+intra+temporal filter	Loop filter set #1 + Rdo + Intra prediction + Temporal filter	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_2.cfg + --EncNnlfOpt=1 + --NnlfSet1Multiframe=1	P	Y
Sr	Super-resolution	encoder_xxx_vtm.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	P	N
set0+sr	Loop filter set #0+Super-resolution	encoder_xxx_vtm.cfg + nn-based/NnlfOption_1.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	N	N
set1+sr	Loop filter set #1+Super-resolution	encoder_xxx_vtm.cfg + nn-based/NnlfOption_2.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	N	N
sr+intra	Super-resolution +Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	P	Y
set0+sr+intra	Loop filter set #0+Super-resolution+Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_1.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	N	N
set1+sr+intra	Loop filter set #01+Super-resolution+Intra prediction	nn-based/encoder_xxx_nnvc.cfg + nn-based/NnlfOption_2.cfg + nn-based/nnsr.cfg + nn-based/nnsr_classAx.cfg	N	N
Pf	Adaptive post-filters	encoder_xxx_vtm.cfg + nn-based/nnpf_int16.cfg	P	P
Intra+pf	Intra prediction + adaptive post-filter	nn-based/encoder_xxx_nnvc.cfg + nn-based/nnpf_int16.cfg	P	N

Recommendations

The AHG recommended to:

Continue to develop NNVC software.
Improve the software documentation.
Encourage people to report all (potential) bugs that they are finding using GitLab Issues functionality https://vcgit.hhi.fraunhofer.de/jvet-ahg-nnvc/VVCSoftware_VTM/-/issues
Encourage people to submit merge requests fixing identified bugs.

It was commented that at some point it might again be useful to test combinations of ECM and NNVC tools.

Project development (23)

AHG1: Deployment and advertisement of standards (3)

Contributions in this area were discussed at 1710–1750 on Thursday 19 Oct. 2023 (chaired by JRO).

ARCHIVE :

JVET-AF0014-v1.docx NNVC-5.0_VTM_vs_NNVC-6.0_VTM_Tencent.xlsm NNVC-6.0_HOP_vs_NNVC-6.1_HOP.xlsm NNVC-6.0_vs_NNVC-6.0_HOP.xlsm NNVC-6.0_VTM_vs_NNVC-6.0_HOP.xlsm NNVC-6.0_VTM_vs_NNVC-6.0_VTM_HOP_Tencent.xlsm

Decisions

merged

The AHG recommended to: Continue to develop NNVC software. Improve the software documentation. Encourage people to report all (potential) bugs that they are finding using GitLab Issues functionality https://vcgit.hhi.fraunhofer.de/jvet-ahg-nnvc/VVCSoftware_VTM/-/issues Encourage people to submit merge requests fixing identified bugs. It was commented that at some point it might again be useful to test combinations of ECM and NNVC tools. Project development (23)

Citation