Single-label activity classification on Breakfast
90.27AccuracyTranS4mer
Evaluation results
| Method | Links | |
|---|---|---|
| TranS4merPretraining Data=495K2022.12 | 90.27 | |
| Transformer w/ KB TransferLong-term Model=Transformer w/ KB Transfer, Segment Model=TimeSformer, Pretraining Supervision=Unsupervised: distant supervision (ours), Pretraining Dataset=HT100M2022.01 | 89.9 | |
| Dist.Sup.Pretraining Data=136M2022.12 | 89.9 | |
| Basic TransformerLong-term Model=Basic Transformer, Segment Model=TimeSformer, Pretraining Supervision=Unsupervised: distant supervision (ours), Pretraining Dataset=HT100M2022.01 | 88.7 | |
| ViS4merPretraining Data=495K2022.12 | 88.17 | |
| Basic TransformerLong-term Model=Basic Transformer, Segment Model=TimeSformer [8], Pretraining Supervision=Unsupervised: k-means on ASR, Pretraining Dataset=HT100M2022.01 | 81.4 | |
| Basic TransformerLong-term Model=Basic Transformer, Segment Model=TimeSformer [8], Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics2022.01 | 81.1 | |
| Basic TransformerLong-term Model=Basic Transformer, Segment Model=SlowFast [17], Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics2022.01 | 76.1 | |
| GHRMLong-term Model=GHRM [69], Segment Model=I3D [11], Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics2022.01 | 75.5 | |
| GHRMPretraining Data=306K2022.12 | 75.5 | |
| Basic TransformerLong-term Model=Basic Transformer, Segment Model=S3D [39], Pretraining Supervision=Unsupervised: MIL-NCE on ASR, Pretraining Dataset=HT100M2022.01 | 74.4 | |
| TimeceptionLong-term Model=Timeception [25], Segment Model=3D-ResNet [62], Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics2022.01 | 71.3 | |
| VideoGraphLong-term Model=VideoGraph [26], Segment Model=I3D [11], Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics2022.01 | 69.5 | |
| VideoGraphBackbone=I3D2019.05 | 69.45 | |
| TimeceptionBackbone=I3D2019.05 | 67.07 | |
| ActionVLADBackbone=I3D2019.05 | 65.48 | |
| VideoGraphBackbone=ResNet-1522019.05 | 59.12 | |
| I3DBackbone=I3D2019.05 | 58.61 | |
| TimeceptionBackbone=ResNet-1522019.05 | 57.75 | |
| ActionVLADBackbone=ResNet-1522019.05 | 55.49 | |
| ResNet-152Backbone=ResNet-1522019.05 | 41.13 |