Lip Reading Classification on LRW-1000 cropped mouth regions (test)
0.466Top-1 AccuracyEnsemble
Evaluation Results
| Method | Links | |
|---|---|---|
| EnsembleData version=Cropped mouth, Ensemble=true2020.07 | 0.466 | |
| ResNet-18 + MS-TCN - Student 1Data version=Cropped mouth, Role=Student 1, Distillation iteration=1, Training Strategy=Born-Again Distillation2020.07 | 0.453 | |
| ResNet-18 + BGRU + CutoutData version=Full face2020.07 | 0.452 | |
| ResNet-18 + MS-TCN - Student 2Data version=Cropped mouth, Role=Student 2, Distillation iteration=2, Training Strategy=Born-Again Distillation2020.07 | 0.447 | |
| ResNet-18 + MS-TCN - TeacherData version=Cropped mouth, Role=Teacher, Training Strategy=Born-Again Distillation2020.07 | 0.432 | |
| ResNet-18 + MS-TCNData version=Cropped mouth2020.07 | 0.414 | |
| ResNet-18 + BGRUData version=Cropped mouth2020.07 | 0.386 | |
| ResNet34 + BLSTMData version=Cropped mouth2020.07 | 0.382 | |
| ResNet34 + DenseNet52 + ConvLSTMData version=Cropped mouth2020.07 | 0.369 |