JVET-AN0361 AHG17: Report on CfE testing [M. Wien, F. Bossen, Ch. Lehmann, V. Baroncini]

Initial results presented Tue 7 Oct. 1425.

Test on Compression capability

Examples from SDR UHD

Examples SDR HD

LB

HDR 4K

HDR 8K

Gaming

It was commented that gaming content is somewhat more difficult to view, and for gaming DCR might have been better to know how it should look like.

UGC

From the compression efficiency test case, the following initial conclusions can be drawn:

• Technically, all submissions are based on VTM, ECM, and/or NNVC, or variants thereof with modifications of encoder or some modification of coding tools (no new coding tools were included)
• Majority of cases (roughly 2/3) where at least one submission has non-overlapping confidence interval with VTM
• In a number of cases, submissions are better in quality than the ECM16 comparison point
• Some proposals show clear improvement in compression versus VTM with substantially less encoder/decoder complexity than ECM
• In various cases (including challenging ones), NN based technology showed clear visual benefit (more than PSNR wise)
• Clearly larger difference for 8K HDR, and less difference for gaming (average rate saving per category?)

From the encoding run time test case, the following initial conclusions can be drawn:

• Significant number of cases (roughly 1/2 for 1x, more for 0.2x) where at least one submission has non-overlapping confidence interval with VTM
• Even if the encoder of a submission is faster than VTM encoder, the decoder may be slower (where one submission comes with only 35% increase).