notesum.ai

Published at December 10

A Generative Victim Model for Segmentation

cs.CV

Released Date: December 10, 2024

Authors: Aixuan Li¹, Jing Zhang¹, Jiawei Shi¹, Yiran Zhong¹, Yuchao Dai¹

Aff.: ¹University

Arxiv: http://arxiv.org/pdf/2412.07274v1

Refer to caption

Victim	ViT [65]				PVTv2 [66]				R50 [67]				Swin [68]				Vgg [69]
	$\mathcal{M}\uparrow$	$CC\downarrow$	$S_{\alpha}\downarrow$	$E_{\xi}\downarrow$	$\mathcal{M}\uparrow$	$CC\downarrow$	$S_{\alpha}\downarrow$	$E_{\xi}\downarrow$	$\mathcal{M}\uparrow$	$CC\downarrow$	$S_{\alpha}\downarrow$	$E_{\xi}\downarrow$	$\mathcal{M}\uparrow$	$CC\downarrow$	$S_{\alpha}\downarrow$	$E_{\xi}\downarrow$	$\mathcal{M}\uparrow$	$CC\downarrow$	$S_{\alpha}\downarrow$	$E_{\xi}\downarrow$
Baseline	.026	.781	.852	.924	.024	.797	.866	.932	.041	.681	.792	.864	.029	.759	.841	.916	.048	.644	.764	.836
MI-FGSM [75]	.264	.287	.497	.491	.090	.562	.718	.766	.105	.451	.654	.713	.074	.574	.730	.798	.134	.368	.594	.661
$\text{DI}^{2}\text{-FGSM}$ [76]	.080	.634	.757	.792	.042	.720	.821	.876	.056	.628	.764	.828	.040	.703	.811	.881	.064	.592	.737	.804
ILA [77]	.121	.480	.654	.701	.043	.704	.808	.874	.064	.580	.732	.804	.043	.687	.796	.876	.084	.504	.682	.753
NAA [78]	.048	.699	.788	.865	.037	.741	.827	.895	.053	.631	.760	.830	.038	.719	.812	.888	.074	.540	.701	.773
RAP [43]	.068	.536	.691	.802	.046	.665	.777	.868	.066	.544	.705	.794	.047	.645	.765	.862	.083	.468	.658	.749
Ours	.069	.570	.708	.802	.090	.497	.665	.756	.123	.364	.596	.670	.088	.482	.658	.756	.160	.270	.526	.617