notesum.ai

Published at November 26

HEIE: MLLM-Based Hierarchical Explainable AIGC Image Implausibility Evaluator

cs.CV

cs.AI

Released Date: November 26, 2024

Authors: Fan Yang¹, Ru Zhen², Jianing Wang³, Yanhao Zhang², Haoxiang Chen⁴, Haonan Lu², Sicheng Zhao¹, Guiguang Ding¹

Aff.: ¹Tsinghua University; ²OPPO AI Center; ³Peking University; ⁴Fudan University

Arxiv: http://arxiv.org/abs/2411.17261v1

Refer to caption

	Heatmap Prediction						Plausibility Score
	MSE (All Data)↓	MSE (GT=0)↓	KLD↓	CC↑	SIM↑	AUC-Judd↑	PLCC↑	SRCC↑
PickScore (off-the-shelf)	-	-	-	-	-	-	0.010	0.028
EVA-CLIP encoder (fine-tuned)	0.01614	0.00512	2.835	0.350	0.082	0.549	0.157	0.143
CLIP encoder (fine-tuned)	0.01437	0.00425	2.462	0.251	0.122	0.747	0.390	0.378
RAHF (multi-head)	0.01216	0.00141	1.971	0.425	0.302	0.877	0.666	0.654
RAHF (augmented prompt)	0.00920	0.00095	1.652	0.556	0.409	0.913	0.693	0.681
HEIE (ours)	0.00825	0.00014	1.634	0.574	0.417	0.915	0.697	0.683