Table 4 Ablation study of the proposed metric-learning strategy on the GazeCapture dataset
From: EM-Gaze: eye context correlation and metric learning for gaze estimation
Method | Phone | Tablet | ||
|---|---|---|---|---|
Error ↓ (cm) | Accuracy ↑ (%) | Error ↓ (cm) | Accuracy ↑ (%) | |
iTracker w/o Llmc | 2.11 | 51.9 | 3.41 | 65.8 |
iTracker w/ Llmc | 2.08 | 63.4 | 3.30 | 79.0 |
AFF-Net w/o Llmc | 1.63 | 57.1 | 2.35 | 70.1 |
AFF-Net w/ Llmc | 1.65 | 68.4 | 2.46 | 83.2 |
EM-Gaze (ours) w/o Llmc | 1.59 | 57.9 | 2.22 | 71.0 |
EM-Gaze (ours) w/ Llmc | 1.57 | 69.2 | 2.21 | 85.1 |