Action Recognition on UCF101 (1)
95.6AccuracyKinetics pretrained I3D
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Kinetics pretrained I3DModel Type=Multi-frame, Pre-training=Kinetics2016.12 | 95.6 | — | — | — | |
| ActionFlowNet (UCF101)Model Type=Multi-frame, Pre-training=FlowNet part pretrained on UCF1012016.12 | 83.9 | — | — | — | |
| Sports-1M pretrained C3DModel Type=Multi-frame, Pre-training=Sports-1M2016.12 | 82.3 | — | — | — | |
| Multi-frame FlowNet fine-tuneModel Type=Multi-frame, Initialization=Pretrained on UCF101 for optical flow2016.12 | 80.8 | — | — | — | |
| ImageNet pretrained ResNet-18Model Type=Two-frame, Initialization=ImageNet2016.12 | 80.7 | — | — | — | |
| DPCBackbone=3D ResNet-34†, Params=33 x 10^6, 2D-CNN=false, Pre-Training=Kinetics4002020.03 | 75.7 | — | — | — | |
| TCEBackbone=2D ResNet-50, Params=23 x 10^6, 2D-CNN=true, Pre-Training=Kinetics4002020.03 | 71.2 | — | — | — | |
| ActionFlowNet-2F (FlCh+UCF101)Model Type=Two-frame, Pre-training=FlowNet part pretrained on FlyingChairs and UCF1012016.12 | 71 | — | — | — | |
| ActionFlowNet-2F (UCF101)Model Type=Two-frame, Pre-training=FlowNet part pretrained on UCF1012016.12 | 70 | — | — | — | |
| StackedModel Type=Two-frame, Initialization=FlowNet pretrained on UCF1012016.12 | 69.6 | — | — | — | |
| TCEBackbone=2D ResNet-18, Params=11 x 10^6, 2D-CNN=true, Pre-Training=Kinetics4002020.03 | 68.8 | — | — | — | |
| DPCBackbone=3D ResNet-18†, Params=14 x 10^6, 2D-CNN=false, Pre-Training=Kinetics4002020.03 | 68.2 | — | — | — | |
| TCEBackbone=2D ResNet-18, Params=11 x 10^6, 2D-CNN=true, Pre-Training=UCF1012020.03 | 68.2 | — | — | — | |
| FlowNet fine-tuneModel Type=Two-frame, Initialization=Pretrained on UCF101 for optical flow2016.12 | 66 | — | — | — | |
| 3DCubicPuzzlesBackbone=3D ResNet-18, Params=34 x 10^6, 2D-CNN=false, Pre-Training=Kinetics4002020.03 | 65.8 | — | — | — | |
| Video Clip OrderingBackbone=R3D, Params=14 x 10^6, 2D-CNN=false, Pre-Training=UCF1012020.03 | 64.9 | — | — | — | |
| Skip-ClipBackbone=3D ResNet-18, Params=34 x 10^6, 2D-CNN=false, Pre-Training=UCF1012020.03 | 64.4 | — | — | — | |
| 3DRotNetBackbone=3D ResNet-18, Params=34 x 10^6, 2D-CNN=false, Pre-Training=Kinetics4002020.03 | 62.9 | — | — | — | |
| DPCBackbone=3D ResNet-18†, Params=14 x 10^6, 2D-CNN=false, Pre-Training=UCF1012020.03 | 60.6 | — | — | — | |
| OPNBackbone=VGG-M-2048, Params=8.6 x 10^6, 2D-CNN=true, Pre-Training=UCF1012020.03 | 59.8 | — | — | — | |
| Arrow of timeBackbone=AlexNet, Params=61 x 10^6, 2D-CNN=true, Pre-Training=UCF1012020.03 | 55.3 | — | — | — | |
| CMCBackbone=CaffeNet x2*, Params=58 x 10^6 x 2, 2D-CNN=true, Pre-Training=UCF1012020.03 | 55.3 | — | — | — | |
| VideoGANBackbone=C3D, Params=11 x 10^6, 2D-CNN=false, Pre-Training=UCF1012020.03 | 52.1 | — | — | — | |
| ScratchModel Type=Two-frame, Backbone=ResNet-18, Initialization=Random2016.12 | 51.3 | — | — | — | |
| BridgeFormerEvaluation Protocol=Zero-shot, Task=Video-to-text retrieval2022.01 | 51.1 | — | — | — | |
| Shuffle and LearnBackbone=AlexNet, Params=61 x 10^6, 2D-CNN=true, Pre-Training=UCF1012020.03 | 50.9 | — | — | — | |
| Motion & AppearanceBackbone=C3D, Params=11 x 10^6, 2D-CNN=false, Pre-Training=UCF1012020.03 | 48.6 | — | — | — | |
| FrozenEvaluation Protocol=Zero-shot, Task=Video-to-text retrieval2022.01 | 45.4 | — | — | — | |
| ClipBertEvaluation Protocol=Zero-shot, Task=Video-to-text retrieval2022.01 | 27.5 | — | — | — | |
| ActionVLADPre-train dataset=ImageNet2019.08 | — | — | — | 92.7 | |
| C3DPre-train dataset=sports-1M, citation=[30]2019.08 | — | 82.3 | — | — | |
| C3DPre-train dataset=sports-1M, citation=[31]2019.08 | — | 85.8 | — | — | |
| CCS + I3DPre-train dataset=ImageNet2019.08 | — | 86.7 | 87.1 | 93.8 | |
| CCS + TSNPre-train dataset=ImageNet2019.08 | — | 87.2 | 87.4 | 95.3 | |
| CCS + TSNPre-train dataset=ImageNet+Kinetics2019.08 | — | 94.2 | 95 | 97.4 | |
| ConvNet fusionPre-train dataset=ImageNet2019.08 | — | 82.6 | 86.2 | 90.6 | |
| ConvNets+LSTMPre-train dataset=ImageNet2019.08 | — | 68.2 | — | — | |
| DTPPPre-train dataset=ImageNet2019.08 | — | 89.7 | 89.1 | 94.9 | |
| I3DPre-train dataset=ImageNet2019.08 | — | 84.5 | 90.6 | 93.4 | |
| R(2+1)DPre-train dataset=sports-1M2019.08 | — | 93.6 | 93.3 | 95 | |
| R(2+1)DPre-train dataset=ImageNet+Kinetics2019.08 | — | 96.8 | 95.5 | 97.3 | |
| ST-resNetPre-train dataset=ImageNet2019.08 | — | 82.3 | 79.1 | 93.4 | |
| TLE+Two-streamPre-train dataset=ImageNet2019.08 | — | — | — | 95.6 | |
| TSNPre-train dataset=ImageNet2019.08 | — | 85.7 | 87.9 | 93.5 | |
| TSNPre-train dataset=ImageNet+Kinetics2019.08 | — | 91.1 | 95.2 | 97 | |
| Two-stream NetworkPre-train dataset=ImageNet2019.08 | — | 73 | 83.7 | 88 |