Figure 9.
A two-panel plot compares P S N R against b p p for U V G and M C L-J C V across H M 16.26, V T M 13.2, V C T and six learned coding methods.The U V G panel uses b p p values from 0.02 to 0.14 and P S N R values from 34.0 to 38.0. H M 16.26 rises from about 34.4 at 0.02 b p p to about 38.0 near 0.135 b p p. V T M 13.2 rises from about 35.1 at 0.02 b p p to about 38.0 near 0.10 b p p. D V C rises from about 34.0 near 0.046 b p p to about 37.2 near 0.128 b p p. D V C pro rises from about 34.0 near 0.035 b p p to about 37.0 near 0.101 b p p. D C V C rises from about 34.0 near 0.028 b p p to about 37.4 near 0.103 b p p. C A N F-V C rises from about 34.0 near 0.027 b p p to about 37.6 near 0.102 b p p. D H V C rises from about 34.8 near 0.034 b p p to about 37.4 near 0.095 b p p. V C T rises from about 35.0 near 0.032 b p p to about 38.0 near 0.135 b p p. A T E M-Hybrid C A rises from about 35.1 near 0.029 b p p to about 38.0 near 0.11 b p p. The M C L-J C V panel uses b p p values from 0.04 to 0.20 and P S N R values from 35.0 to 39.0. H M 16.26 rises from about 35.8 at 0.04 b p p to about 38.8 near 0.20 b p p. V T M 13.2 rises from about 36.1 at 0.04 b p p to about 39.0 near 0.155 b p p. D V C rises from about 35.0 near 0.072 b p p to about 37.5 near 0.155 b p p. D V C pro rises from about 35.1 near 0.059 b p p to about 37.4 near 0.128 b p p. D C V C rises from about 35.5 near 0.049 b p p to about 37.7 near 0.131 b p p. C A N F-V C rises from about 35.6 near 0.048 b p p to about 38.0 near 0.143 b p p. D H V C rises from about 35.7 near 0.042 b p p to about 38.9 near 0.176 b p p. V C T rises from about 35.8 near 0.049 b p p to about 38.2 near 0.147 b p p. A T E M-Hybrid C A rises from about 35.9 near 0.044 b p p to about 39.0 near 0.20 b p p.

Rate-distortion curve of VCT baseline and best performing Conv-Attn hybrid mixer in ATEM for the UVG and MCL-JCV data sets. Traditional codecs [HM (JVET/HM, 2024) and VTM (VCGIT, 2024)] are indicated by dashed lines for reference. For broader context, pioneering methods DVC (Lu et al., 2018), DVC-Pro (Lu et al., 2020), DCVC (Li et al., 2021), DHVC (Lu et al., 2023), and CANF-VC (Ho et al., 2022), proposed before and after VCT are also included for broader comparison. The performance metrics for HM-16.26 are adopted from the DHVC paper (Lu et al., 2023) and VTM-13.2 results are adopted from the benchmarks reported in DCVC-TCM (Sheng et al., 2021)

Source: Authors’ own work

or Create an Account

Close subscription notice
Close access options