JVET-AK0221 [AHG17] JVET CTC and non-CTC sequences performance analysis [E. Alshina, J. Pardo, J. Sauer, T. Solovyev (Huawei)]

This contribution analyzes performance of VTM and ECM for sequences from JVET common test conditions and outside of those, including sequences studied in AhG17 and some sequences from new proposals. Sequences are classified based on the difficulty of encoding. The most challenging sequences are recommended to be in the focus of JVET, studying the needs of the evidence for next generation video coding.

The point of assuming transparent quality at 39 dB PSNR may be too high for more complex sequences. Comparison with the visual test results shows that MOS 8-9 is often achieved by much lower rates.

Investigation is made also based on typical bit rates used for encoding certain types of material in certain application domains. It was commented that for 4K HDR the assumption of up to 85 Mbit/s goes too high – typically not more than double rate would be used than for SDR.

The approach of selecting target rates for testing based on typical application scenario requirements is deemed to be useful.

It was commented that in practice rate control would be used to reach certain target rates. With constant QP, many bits might be spent in vain by coding certain areas with better quality.

It was commented that other metrics than PSNR might be used.

The test should not only focus on content that is difficult to code. However, content that is more in the “normal” range would be relevant as well. Of high importance is the suitability for subjective viewing as well.