Image Safety Assessment on HADES (test) benchmark leaderboard · SOTA2 Research

Image Safety Assessment on HADES (test)

64.815Ensemble ScoreMGSIA

Evaluation Results

Method	Links
MGSIA2026.06		64.815	83.352	60.575	80.32	77.16
MGSIA2026.06		60.208	86.225	98.664	83.293	54.887
AttackVLM-it2026.06		52.995	74.226	60.145	78.727	76.917
Target Text Embedding2026.06		47.116	70.363	56.18	77.503	72.181
AttackVLM-ii2026.06		43.91	64.082	55.986	78.64	74.62
Gaussian Blur2026.06		43.061	65.95	63.601	78.907	65.133
Padding2026.06		41.492	63.73	61.323	78.307	72.793
Chain of Attack2026.06		41.122	63.63	59.155	78.04	71.82
Gaussian Noise2026.06		40.276	62.411	59.834	77.733	68.667
Benign2026.06		40.149	62.348	59.55	78.107	71.283
Target Text Overlay2026.06		40.092	64.566	57.456	79.193	68.067
Bounding Box2026.06		37.807	60.748	57.523	78.347	72.827
Target Text Embedding2026.06		33.332	62.86	89.094	48.78	18.92
Benign2026.06		32.143	46.459	81.837	27.92	11.2
Padding2026.06		30.201	44.873	81.266	30.02	11.687
AttackVLM-it2026.06		30.014	47.23	81.779	34.84	10.993
AttackVLM-ii2026.06		29.587	46.157	81.336	32.773	10.987
Gaussian Noise2026.06		29.497	44.964	82.068	33.073	11.273
Bounding Box2026.06		29.497	46.144	79.762	32.667	16.7
Target Text Overlay2026.06		25.902	44.731	80.189	31.58	10.66
Chain of Attack2026.06		23.845	41.171	80.571	32.007	11.667
Gaussian Blur2026.06		19.594	48.708	83.514	26.367	5.207