Action Quality Assessment on MTL-AQA
0.9315Spearman CorrelationResNet34-(2+1)D-WD
Evaluation Results
| Method | Links | |
|---|---|---|
| ResNet34-(2+1)D-WDBackbone=ResNet34-(2+1)D, Aggregation Technique=WD, Input Frames=32 frames2021.02 | 0.9315 | |
| MUSDLImplementation=Proposed Multi-path Uncertainty-aware Score Distribution Learning2020.06 | 0.9273 | |
| MUSDL2021.02 | 0.9273 | |
| C3D-WDBackbone=C3D, Aggregation Technique=WD, Input Frames=16 frames, Pretraining=Sports-1M [8]2021.02 | 0.9223 | |
| C3D-AVG-MTLLearning Strategy=Multi-Task Learning (MTL)2020.06 | 0.9044 | |
| C3D-AVG-MTL2021.02 | 0.9044 | |
| C3D-AVG-STLLearning Strategy=Single-Task Learning (STL)2020.06 | 0.896 | |
| C3D-AVG-STL2021.02 | 0.896 | |
| ResNet50-3D-WDBackbone=ResNet50-3D, Aggregation Technique=WD, Input Frames=16 frames2021.02 | 0.8935 | |
| RegressionImplementation=Authors' baseline regression2020.06 | 0.8905 | |
| USDL-Regression2021.02 | 0.8905 | |
| MSCADC-MTLLearning Strategy=Multi-Task Learning (MTL)2020.06 | 0.8612 | |
| MSCADC-MTL2021.02 | 0.8612 | |
| C3D-LSTMBackbone=C3D, Temporal Model=LSTM2020.06 | 0.8489 | |
| C3D-LSTM2021.02 | 0.8489 | |
| MSCADC-STLLearning Strategy=Single-Task Learning (STL)2020.06 | 0.8472 | |
| MSCADC-STL2021.02 | 0.8472 | |
| SSL CATE Effect-Affinity AssessmentEvaluation protocol=Linear probing, Number of sampled frames=162024.01 | 0.7936 | |
| Motion Disentanglingabbreviation=MD2022.02 | 0.7763 | |
| SSL Motion DisentanglerEvaluation protocol=Linear probing, Number of sampled frames=162024.01 | 0.7763 | |
| C3D-SVRBackbone=C3D, Regression Head=SVR2020.06 | 0.7716 | |
| C3D-SVR2021.02 | 0.7716 | |
| SSL SoTA2022.02 | 0.77 | |
| SSL Action AlignmentEvaluation protocol=Linear probing, Number of sampled frames=162024.01 | 0.77 | |
| ResNet101-3D-AVGBackbone=ResNet101-3D, Aggregation Technique=AVG, Input Frames=16 frames2021.02 | 0.6633 | |
| Ours baselinepretraining=supervised dive-classification2022.02 | 0.5665 | |
| Pose+DCT2020.06 | 0.2682 | |
| Pose+DCT2021.02 | 0.2682 |