Table 2 Performance of visual saliency models and baselines.

Model	NSS	Authors	Improvement (%)
Baselines
Fixation map	0.709		100.00
Central bias	0.014	Tatler¹⁴	0.00
Single observer	0.176		23.32
Meaning map	0.382	Henderson & Hayes¹⁵	52.97
Models
SALICON	0.462	Jiang et al.²¹	64.45
SalGAN	0.434	Pan et al.²²	60.42
DeepGazeIIE	0.411	Linardos et al.²³	57.15
DeepGazeII	0.408	Kümmerer et al.³⁵	56.78
QSS	0.350	Schauerte & Stiefelhagen⁴⁵	48.37
IMSIG	0.342	Hou et al.⁴⁶	47.20
DeepGazeI	0.339	Kümmerer et al.⁴⁷	46.83
DVA	0.307	Hou & Zhang⁴⁸	42.13
SSR	0.280	Seo & Milanfar⁴⁹	38.35
SAM	0.279	Cornia et al.²⁰	38.10
ICF	0.269	Kümmerer et al.⁵⁰	36.70
AIM	0.255	Bruce & Tsotsos⁵¹	34.75
IKN	0.206	Itti et al.¹	27.66
RARE2012	0.200	Riche et al.⁵²	26.79
BMS	0.194	Zhang & Sclaroff⁵³	25.88
CAS	0.172	Goferman et al.⁵⁴	22.71
GBVS	0.171	Harel et al.⁵⁵	22.65
SUN	0.166	Zhang et al.⁵⁶	21.85
FES	0.060	Rezazadegan Tavakoli et al.⁵⁷	6.72
LDS	0.043	Fang et al.⁵⁸	4.29
CVS	−0.076	Erdem & Erdem⁵⁹	−12.86

Model performance (NSS, Normalized Scanpath Saliency score) for baselines and models are given as rows. Negative numbers denote worse than chance performance. The Improvement column denotes relative performance between central bias (0%) and fixation map (100%) in % NSS.

Quick links

Search