Skip to main content

Table 1 Gaze regression and classification results on the GazeCapture dataset. For SAGE [3] and TAT [52] methods, classification results are unavailable from publicly accessible contents. EM-Gaze outperforms the alternative methods under gaze-regression error and Top-1 classification accuracy metrics

From: EM-Gaze: eye context correlation and metric learning for gaze estimation

Method

Phone

Tablet

Error ↓ (cm)

Accuracy ↑ (%)

Error ↓ (cm)

Accuracy ↑ (%)

iTracker [2]

2.06

51.9

3.22

65.8

SAGE [3]

1.78

-

2.72

-

TAT [52]

1.77

-

2.66

-

AFF-Net [4]

1.62

57.1

2.30

70.1

EM-Gaze (ours)

1.57

69.2

2.21

85.1