Frame-level hate localization on MHC (test) benchmark leaderboard · SOTA2 Research

Frame-level hate localization on MHC (test)

0.82AccuracyWang et al.

Evaluation Results

Method	Links
Wang et al.2026.02		0.82	—	—	0.81	0.76	0.87	0.68
HVGuard2026.02		0.809	—	—	0.6646	0.4556	0.4722	0.5
Yue et al.2026.02		0.78	—	—	—	0.77	0.8	0.77
MHCL2026.02		0.7498	—	—	0.7317	0.6153	0.675	0.6783
LELA2026.02		0.7124	0.7227	0.6733	0.6923	0.6568	0.8217	0.5387
HTMM2026.02		0.6861	—	—	0.7456	0.6817	0.6875	0.6597
Lin et al.2026.02		0.6174	0.6147	0.5626	0.5849	0.5637	0.7233	0.4813
LAVAD2026.02		0.5833	0.5865	0.6302	0.5799	0.5344	0.5671	0.6926
ZS ImageBind2026.02		0.5391	0.5135	0.5753	0.4756	0.6392	0.5074	0.5014
LLAVA-1.52026.02		0.5342	0.5319	0.5438	0.5062	0.4974	0.5766	0.4694
ZS-CLIP2026.02		0.5021	0.5181	0.5449	0.6395	0.5794	0.5371	0.5289