Gaze Estimation on MPIIFaceGaze DM (test)
4.53Angular Error (d)Ours (ViT-B)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ours (ViT-B)Backbone=ViT-B, Model Size=86.6M, Training Datasets={DX, DC, DN}2026.03 | 4.53 | 3.32 | 2.45 | |
| Ours (MobileNet)Backbone=MobileNet, Model Size=3.8M, Training Datasets={DX, DC, DN}2026.03 | 4.83 | 3.5 | 2.67 | |
| GLABackbone=ResNet-18, Model Size=11.7M, Training Datasets={DX, DC, D360}2026.03 | 4.89 | — | — | |
| UniGaze-H-14-JointBackbone=ViT-H, Model Size=632M, Training Datasets=5 datasets2026.03 | 5.08 | 3.86 | 2.54 | |
| Ours (ViT-B)Model Size=86.6M, Backbone=ViT-B, Training Dataset=DX (ETH-XGaze)2026.03 | 5.15 | 3.67 | 2.97 | |
| UniGaze-B-16-JointBackbone=ViT-B, Model Size=86.6M, Training Datasets=5 datasets2026.03 | 5.35 | 3.95 | 2.78 | |
| Ours (MobileNet)Model Size=3.8M, Backbone=MobileNet, Training Dataset=DX (ETH-XGaze)2026.03 | 5.56 | 3.94 | 3.2 | |
| UniGaze-H-14-CrossXModel Size=632M, Backbone=ViT-H, Training Dataset=DX (ETH-XGaze)2026.03 | 5.77 | 4.72 | 2.51 | |
| UniGaze-B-16Model Size=86.6M, Backbone=ViT-B, Training Dataset=DX (ETH-XGaze)2026.03 | 6.21 | — | — | |
| PureGazeModel Size=31M, Backbone=ResNet-50, Training Dataset=DX (ETH-XGaze)2026.03 | 6.79 | 4.86 | 3.87 | |
| GLAModel Size=11.7M, Backbone=ResNet-18, Training Dataset=DX (ETH-XGaze)2026.03 | 6.83 | — | — | |
| ETH-XGazeModel Size=25.6M, Backbone=ResNet-50, Training Dataset=DX (ETH-XGaze)2026.03 | 6.88 | 5.1 | 3.73 |