Gaze Estimation on Gaze360 (test)
10.41MAE (All 360°)L2CS-Net
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| L2CS-Net#Params=23.8M2026.05 | 10.41 | — | — | — | 19.01 | — | — | — | 15.33 | 18.88 | 22.82 | |
| CA-Ne#Params=34.1M2026.05 | 11.2 | — | — | — | — | — | — | — | — | — | — | |
| ST-WSGE#Params=27.5M2026.05 | 11.58 | — | — | — | 17.08 | — | — | — | 13.71 | 16.2 | 21.33 | |
| DHECA-SuperGaze-TModel Category=Temporal2025.05 | 11.89 | 10.23 | 9.71 | — | — | — | 17.93 | — | — | — | — | |
| Gazetr#Params=11.4M2026.05 | 12.11 | — | — | — | 16.57 | — | — | — | 13.67 | 16.19 | 19.86 | |
| RT-Gene#Params=82.0M2026.05 | 12.26 | — | — | — | 13.67 | — | — | — | 13.02 | 13.42 | 14.58 | |
| MSA+SeqModel Category=Temporal2025.05 | 12.48 | 10.68 | 10.15 | — | — | — | 18.97 | — | — | — | — | |
| MSA+SeqTraining Data=G360V2025.02 | 12.5 | 10.7 | — | — | — | — | 19 | — | — | — | — | |
| MCGazeTraining Data=G360V2025.02 | 12.96 | 10.74 | 10.02 | — | — | — | — | — | — | — | — | |
| FullFace#Params=197M2026.05 | 12.99 | — | — | — | 14.93 | — | — | — | 13.61 | 14.91 | 16.26 | |
| MCGazeModel Category=Temporal, Utilization of eye/face annotations=True2025.05 | 13.01 | 10.99 | 10.62 | — | — | — | 20.22 | — | — | — | — | |
| ST-WSGE (GaT)Training Data=G360I&V+GF2025.02 | 13.19 | 11.34 | 10.84 | — | — | 8.58 | 19.82 | — | — | — | — | |
| Proposed Temporal Model (Gaze360)Training Data=Gaze360, Supervision=Gaze + LAEO Labels, Protocol=Within dataset2021.05 | 13.2 | — | — | — | — | — | — | — | — | — | — | |
| Proposed Temporal Model (Gaze360 + AVA)Training Data=Gaze360 + AVA, Supervision=Gaze + LAEO Labels, Protocol=Within dataset2021.05 | 13.2 | — | — | — | — | — | — | — | — | — | — | |
| Kothari et al.Training Data=G360V2025.02 | 13.2 | — | 10.1 | — | — | — | — | — | — | — | — | |
| Kothari et al.Training Data=G360V+AVA2025.02 | 13.2 | — | 10.2 | — | — | — | — | — | — | — | — | |
| DHECA-SuperGaze-SModel Category=Static2025.05 | 13.33 | 11.67 | 11.45 | — | — | — | 19.29 | — | — | — | — | |
| Pinball LSTMUncertainty Loss=Yes2019.10 | 13.5 | 11.4 | 11.1 | 0.45 | — | — | — | — | — | — | — | |
| Gaze360Training Data=Gaze360, Supervision=Gaze + LAEO Labels, Protocol=Within dataset2021.05 | 13.5 | — | — | — | — | — | — | — | — | — | — | |
| Gaze360Training Data=G360V2025.02 | 13.5 | 11.4 | 11.1 | — | — | — | — | — | — | — | — | |
| Baseline#Params=11.1M2026.05 | 13.59 | — | — | — | 18.19 | — | — | — | 14.8 | 16.76 | 23 | |
| Supervised (GaT)Training Data=G360I&V2025.02 | 13.64 | 11.66 | 11.2 | — | — | 9.1 | 20.74 | — | — | — | — | |
| Pinball LSTMModel Category=Temporal2025.05 | 13.68 | 11.44 | 11.32 | — | — | — | 21.75 | — | — | — | — | |
| Dilated-Net#Params=3.90M2026.05 | 13.73 | — | — | — | 14.99 | — | — | — | 13.83 | 15.04 | 16.11 | |
| CrossGazeModel Category=Static2025.05 | 13.81 | 11.97 | 11.8 | — | — | — | 20.54 | — | — | — | — | |
| MSATraining Data=G360I2025.02 | 13.9 | 12.2 | — | — | — | — | 23.5 | — | — | — | — | |
| MSAModel Category=Static2025.05 | 13.9 | 12.23 | 12.25 | — | — | — | 19.9 | — | — | — | — | |
| MSE LSTMUncertainty Loss=No2019.10 | 14.1 | 12.1 | 11.6 | — | — | — | — | — | — | — | — | |
| MSE+Drop LSTMUncertainty Loss=No, Dropout=Yes2019.10 | 14.1 | 12.1 | 11.6 | 0.31 | — | — | — | — | — | — | — | |
| Crop Aug. LSTMUncertainty Loss=No, Data Augmentation=Crop2019.10 | 14.1 | 11.6 | 11.2 | 0.37 | — | — | — | — | — | — | — | |
| PinBall TRNUncertainty Loss=Yes2019.10 | 14.1 | 11.7 | 11.6 | 0.46 | — | — | — | — | — | — | — | |
| Crop Aug. TRNUncertainty Loss=No, Data Augmentation=Crop2019.10 | 14.2 | 11.5 | 11.4 | 0.39 | — | — | — | — | — | — | — | |
| MSE TRNUncertainty Loss=No2019.10 | 14.3 | 11.8 | 11.8 | — | — | — | — | — | — | — | — | |
| MSE+Drop TRNUncertainty Loss=No, Dropout=Yes2019.10 | 14.3 | 11.8 | 11.8 | 0.31 | — | — | — | — | — | — | — | |
| Kothari et al.Training Data=G360I2025.02 | 15.07 | — | — | — | — | 10.94 | — | — | — | — | — | |
| GazeTR-HybridModel Category=Static2025.05 | 15.29 | 12.88 | 13.06 | — | — | — | 23.94 | — | — | — | — | |
| PinBall StaticUncertainty Loss=Yes2019.10 | 15.6 | 13.4 | 13.2 | 0.42 | — | — | — | — | — | — | — | |
| Gaze360Training Data=G360I2025.02 | 15.6 | 13.4 | — | — | — | 13.2 | — | — | — | — | — | |
| MSE StaticUncertainty Loss=No2019.10 | 15.8 | 13.7 | 13.4 | — | — | — | — | — | — | — | — | |
| MSE+Drop StaticUncertainty Loss=No, Dropout=Yes2019.10 | 15.8 | 13.7 | 13.4 | 0.24 | — | — | — | — | — | — | — | |
| L2CS-NetModel Category=Static2025.05 | 15.81 | 13.12 | 13.14 | — | — | — | 25.49 | — | — | — | — | |
| Pinball StaticModel Category=Static2025.05 | 15.95 | 13.09 | 12.97 | — | — | — | 26.24 | — | — | — | — | |
| Crop Aug. StaticUncertainty Loss=No, Data Augmentation=Crop2019.10 | 16 | 13.2 | 12.6 | 0.37 | — | — | — | — | — | — | — | |
| Hybrid-SAM+LSTMModel Category=Temporal2025.05 | 17.02 | 13.28 | 12.38 | — | — | — | 30.5 | — | — | — | — | |
| Hybrid-SAM+TxModel Category=Temporal2025.05 | 17.32 | 13.44 | 12.45 | — | — | — | 31.31 | — | — | — | — | |
| ST-WSGE (GaT)Training Data=GFIEI&V+GF2025.02 | 21.48 | 20.61 | 20.46 | — | — | 26.55 | 24.61 | — | — | — | — | |
| ST-WSGE (GaT)Training Data=GFIEI&V+MPS+GF2025.02 | 21.59 | 17.02 | 16 | — | — | 13.9 | 38.02 | — | — | — | — | |
| Proposed Temporal Model (CMU Panoptic + AVA)Training Data=CMU Panoptic + AVA, Supervision=LAEO Labels, Protocol=Cross dataset2021.05 | 24.4 | — | — | — | — | — | — | — | — | — | — | |
| Proposed Temporal Model (ETH-XGaze + AVA)Training Data=ETH-XGaze + AVA, Supervision=Gaze + LAEO Labels, Protocol=Cross dataset2021.05 | 25 | — | — | — | — | — | — | — | — | — | — | |
| Kothari et al.Training Data=ETH+AVA2025.02 | 25 | — | — | — | — | 16.9 | — | — | — | — | — | |
| Supervised (GaT)Training Data=GFIEI&V+MPS2025.02 | 25.75 | 20.35 | 19.07 | — | — | 16.35 | 45.19 | — | — | — | — | |
| Proposed Temporal Model (CMU Panoptic)Training Data=CMU Panoptic, Supervision=LAEO Labels, Protocol=Cross dataset2021.05 | 25.9 | — | — | — | — | — | — | — | — | — | — | |
| Proposed Temporal Model (AVA)Training Data=AVA, Supervision=LAEO Labels, Protocol=Cross dataset2021.05 | 26.3 | — | — | — | — | — | — | — | — | — | — | |
| Proposed Temporal Model (GazeCapture + AVA)Training Data=GazeCapture + AVA, Supervision=Gaze + LAEO Labels, Protocol=Cross dataset2021.05 | 27.2 | — | — | — | — | — | — | — | — | — | — | |
| Supervised (GaT)Training Data=GFIEI&V2025.02 | 30.57 | 29.08 | 28.87 | — | — | 33.43 | 35.95 | — | — | — | — | |
| Deep HP2019.10 | 49.3 | 30.7 | 22.7 | — | — | — | — | — | — | — | — | |
| Proposed Temporal Model (ETH-XGaze)Training Data=ETH-XGaze, Supervision=Gaze + LAEO Labels, Protocol=Cross dataset2021.05 | 52.6 | — | — | — | — | — | — | — | — | — | — | |
| Kothari et al.Training Data=ETH2025.02 | 52.6 | — | — | — | — | 30.5 | — | — | — | — | — | |
| Proposed Temporal Model (GazeCapture)Training Data=GazeCapture, Supervision=Gaze + LAEO Labels, Protocol=Cross dataset2021.05 | 58.2 | — | — | — | — | — | — | — | — | — | — | |
| Mean2019.10 | 59 | 40.5 | 19 | — | — | — | — | — | — | — | — | |
| 3DGazeNetTraining Data=ETH, restricted frontal pose=true2025.02 | — | — | 22.1 | — | — | — | — | — | — | — | — | |
| 3DGazeNetTraining Data=ETH+AVA+CMU, restricted frontal pose=true2025.02 | — | — | 17 | — | — | — | — | — | — | — | — | |
| 3DGazeNetTraining Data=ETH+ITWG-MV, restricted frontal pose=true2025.02 | — | — | 15.4 | — | — | — | — | — | — | — | — | |
| Baseline + DSCL#Params=11.1M2026.05 | — | — | — | — | 13.6 | — | — | — | 12.67 | 13.94 | 14.21 | |
| CA-Net2022.03 | — | — | — | — | 12.26 | — | — | — | — | — | — | |
| CA-NetTraining Data=G360I Face2025.02 | — | — | — | — | 11.2 | — | — | — | — | — | — | |
| DilatedTraining Data=G360I Face2025.02 | — | — | — | — | 13.73 | — | — | — | — | — | — | |
| Dilated-Net2022.03 | — | — | — | — | 13.73 | — | — | — | — | — | — | |
| Dilated-Net + DSCL#Params=3.90M2026.05 | — | — | — | — | 14.65 | — | — | — | 13.57 | 14.83 | 15.55 | |
| FullFace2022.03 | — | — | — | — | 14.99 | — | — | — | — | — | — | |
| FullFaceTraining Data=G360I Face2025.02 | — | — | — | — | 14.99 | — | — | — | — | — | — | |
| FullFace + DSCL#Params=197M2026.05 | — | — | — | — | 14.31 | — | — | — | 13.03 | 14.81 | 15.09 | |
| Gaze360 (LSTM)2022.03 | — | — | — | — | 11.4 | — | — | — | — | — | — | |
| GazeTRTraining Data=G360I Face2025.02 | — | — | — | — | 10.62 | — | — | — | — | — | — | |
| Gazetr + DSCL#Params=11.4M2026.05 | — | — | — | — | 15.14 | — | — | — | 13.55 | 14.36 | 17.53 | |
| Holistic MLPModel Type=Landmark-based2026.03 | — | — | — | — | 14.95 | — | — | 12.66 | — | — | — | |
| L2CS-Netbeta (β)=22022.03 | — | — | — | — | 10.54 | — | — | — | — | — | — | |
| L2CS-Netbeta (β)=12022.03 | — | — | — | — | 10.41 | — | — | — | — | — | — | |
| ResNet18Model Type=Image-based2026.03 | — | — | — | — | 12.23 | — | — | — | — | — | — | |
| ResNet50Training Data=G360I Face2025.02 | — | — | — | — | 10.73 | — | — | — | — | — | — | |
| RT-Geneensemble=42022.03 | — | — | — | — | 12.26 | — | — | — | — | — | — | |
| RT-GeneTraining Data=G360I Face2025.02 | — | — | — | — | 12.26 | — | — | — | — | — | — | |
| RT-Gene + DSCL#Params=82.0M2026.05 | — | — | — | — | 13.15 | — | — | — | 12.83 | 13.22 | 13.41 | |
| Siamese MLPModel Type=Landmark-based2026.03 | — | — | — | — | 14.68 | — | — | 12.7 | — | — | — | |
| ST-WSGE (GaT)Training Data=G360I&V+GF2025.02 | — | — | — | — | 10.84 | — | — | — | — | — | — | |
| ST-WSGE + DSCL#Params=27.5M2026.05 | — | — | — | — | 14.79 | — | — | — | 13.05 | 14.21 | 17.1 | |
| Supervised (GaT)Training Data=G360I&V2025.02 | — | — | — | — | 11.2 | — | — | — | — | — | — | |
| XGBoostModel Type=Landmark-based2026.03 | — | — | — | — | 18.81 | — | — | 14.38 | — | — | — |