Audio-driven human video generation on Combined Author dataset + PATS (test) benchmark leaderboard · SOTA2 Research

Audio-driven human video generation on Combined Author dataset + PATS (test)

0.8275SSIMOurs

Evaluation Results

Method	Links
Ours2025.03		0.8275	0.1368	45.41	37.45	527.24	0.2355	0.9485	0.8832	0.3401	0.7454
w/o R2-DiT2025.03		0.8118	0.1399	48.27	37.57	549.01	0.233	0.9428	0.8579	0.3319	0.7248
w/o MT2025.03		0.8023	0.1795	45.75	37.52	629.3	0.2343	0.9424	0.8283	0.3264	0.7376
w/o HPE2025.03		0.7999	0.1416	50.37	35.13	660.13	0.232	0.9319	0.8526	0.3278	0.7331
w/o IA2025.03		0.7626	0.1454	48.69	37.59	575.94	0.232	0.9419	0.8434	0.3376	0.6839
w/o AA2025.03		0.7253	0.1927	63.94	35.11	667.29	0.2309	0.9457	0.8408	0.3395	0.7016
Talk-CNxt2025.03		0.667	0.302	80.49	85.44	1,289.06	0.2187	0.9364	0.7633	0.4127	0.4887
Prob-CNxt2025.03		0.663	0.2966	74.2	62.65	1,420.41	0.226	0.9352	0.8677	0.4642	0.616
Prob-Mimic2025.03		0.5546	0.44	131.25	112.85	1,563.95	0.2297	0.9172	0.8691	0.4365	0.3265
Talk-Mimic2025.03		0.5497	0.4375	136.82	134.02	1,648.92	0.2257	0.8976	0.7465	0.3467	0.419