Hallucination Detection on VQA v2 benchmark leaderboard · SOTA2 Research

Hallucination Detection on VQA v2

0.777ROCP(True)

Evaluation Results

Method	Links
P(True)2026.01		0.777	0.86	0.499	0.915	1.1437	24.299
SEU2026.01		0.772	0.787	0.667	0.868	1.686	1.845
FaithSCAN2026.01		0.752	0.82	0.573	0.858	1.5602	0.605
EmbVar2026.01		0.724	0.773	0.667	0.845	1.4309	0.06
SelfCheckGPT2026.01		0.712	0.851	0.43	0.877	2.165	367.522
Logistic2026.01		0.71	0.76	0.555	0.806	1.5703	0.087
P(True)2026.01		0.703	0.782	0.554	0.777	1.128	14.938
FaithSCAN2026.01		0.695	0.876	0.42	0.884	1.7417	1.191
SelfCheckGPT2026.01		0.693	0.741	0.533	0.755	2.5881	340.683
FaithSCAN2026.01		0.685	0.801	0.494	0.831	1.899	0.741
Logistic2026.01		0.675	0.812	0.411	0.81	1.6895	0.102
P(True)2026.01		0.663	0.772	0.564	0.825	1.1781	19.511
SE2026.01		0.63	0.186	0.425	0.172	2.6328	11.212
PE2026.01		0.624	0.239	0.561	0.225	1.3263	0.002
T-NLL2026.01		0.623	0.265	0.502	0.285	1.2575	0.001
SEU2026.01		0.614	0.805	0.396	0.848	2.1994	1.821
SEU2026.01		0.604	0.666	0.429	0.645	2.4439	1.769
EmbVar2026.01		0.593	0.437	0.481	0.345	2.5046	0.175
Logistic2026.01		0.592	0.797	0.384	0.802	1.8749	0.114
EmbVar2026.01		0.587	0.777	0.483	0.823	2.1946	0.107
SE2026.01		0.565	0.311	0.491	0.297	1.3819	4.913
T-NLL2026.01		0.558	0.725	0.413	0.787	1.5247	0.001
PE2026.01		0.553	0.588	0.433	0.565	1.4914	0.002
PE2026.01		0.549	0.394	0.428	0.305	1.3284	0.002
SE2026.01		0.532	0.208	0.374	0.205	2.0199	13.812
T-NLL2026.01		0.524	0.193	0.312	0.195	1.331	0.001
SelfCheckGPT2026.01		0.502	0.684	0.514	0.657	1.6029	78.613