Action Detection on THUMOS 2014 (test)
51.6mAP (alpha=0.5)G-TAD+P-GCN
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| G-TAD+P-GCNPost-processing=P-GCN2019.11 | 51.6 | 60.4 | 66.4 | — | — | — | 37.6 | 22.9 | |
| BSN+P-GCNPost-processing=P-GCN2019.11 | 49.1 | 57.8 | 63.6 | — | — | — | — | — | |
| TAL-Net2019.11 | 42.8 | 48.5 | 53.2 | — | — | — | 33.8 | 20.8 | |
| G-TAD2019.11 | 40.2 | 47.6 | 54.5 | — | — | — | 30.8 | 23.4 | |
| DBG2019.11 | 39.8 | 49.4 | 57.8 | — | — | — | 30.2 | 21.7 | |
| BMN2019.11 | 38.8 | 47.4 | 56 | — | — | — | 29.7 | 20.5 | |
| MGG2019.11 | 37.4 | 46.8 | 53.9 | — | — | — | 29.5 | 21.3 | |
| BSN2019.11 | 36.9 | 45 | 53.5 | — | — | — | 28.4 | 20 | |
| Two-stream R-C3DFusion strategy=Sum, OHEM=true2019.06 | 36.1 | 43 | 51.2 | 54.7 | 56.9 | — | — | — | |
| Single-stream R-C3DOHEM=true2019.06 | 35.8 | 43.1 | 51.1 | 54.9 | 57.4 | — | — | — | |
| Two-stream R-C3DFusion strategy=Sum2019.06 | 33.4 | 40.6 | 48.9 | 54.2 | 56.6 | — | — | — | |
| Two-stream R-C3DFusion strategy=Concat2019.06 | 33.1 | 40 | 46.9 | 52.2 | 54.5 | — | — | — | |
| CBR2019.11 | 31 | 41.3 | 50.1 | — | — | — | 19.1 | 9.9 | |
| Zhao et al.2019.06 | 29.8 | 41 | 51.9 | 59.4 | 66 | — | — | — | |
| SSN2019.11 | 29.8 | 41 | 51.9 | — | — | — | — | — | |
| SS-TAD2019.11 | 29.2 | — | 45.7 | — | — | — | — | 9.6 | |
| Single-stream R-C3DBuffer type=two-way buffer2019.06 | 28.9 | 35.6 | 44.8 | 51.5 | 54.5 | — | — | — | |
| R-C3DBuffer configuration=two-way buffer2017.03 | 28.9 | 35.6 | 44.8 | 51.5 | 54.5 | — | — | — | |
| Temporal Actionness Grouping (TAG)Backbone=Inception V32017.03 | 28.2 | 39.8 | 48.7 | 57.7 | 64.1 | — | — | — | |
| Single-stream R-C3DBuffer type=one-way buffer2019.06 | 27 | 33.4 | 42.8 | 49.2 | 51.6 | — | — | — | |
| R-C3DBuffer configuration=one-way buffer2017.03 | 27 | 33.4 | 42.8 | 49.2 | 51.6 | — | — | — | |
| Dai et al.2019.06 | 25.6 | 33.3 | — | — | — | — | — | — | |
| TURN-TAP2019.11 | 25.6 | 34.9 | 44.1 | — | — | — | — | — | |
| TCN2019.11 | 25.6 | 33.3 | — | — | — | — | 15.9 | 9 | |
| Shou et al. [41]2019.06 | 23.3 | 29.4 | 40.1 | — | — | — | — | — | |
| CDC2019.11 | 23.3 | 29.4 | 40.1 | — | — | — | 13.1 | 7.9 | |
| Shou et al.2017.03 | 23.3 | 29.4 | 40.1 | — | — | — | — | — | |
| SST2019.11 | 23 | — | — | — | — | — | — | — | |
| Hou et al.2019.11 | 22 | — | 43.7 | — | — | — | — | — | |
| S-CNN2017.03 | 19 | 28.7 | 36.3 | 43.5 | 47.7 | — | — | — | |
| Shou et al.Supervision=Strong2017.03 | 19 | 28.7 | 36.3 | 43.5 | 47.7 | — | — | — | |
| Shou et al. [3]2019.06 | 19 | 28.7 | 36.3 | 43.5 | 47.7 | — | — | — | |
| Shou et al.2017.03 | 19 | 28.7 | 36.3 | 43.5 | 47.7 | — | — | — | |
| Yuan et. al.2017.03 | 18.8 | 26.1 | 33.6 | 42.6 | 51.4 | — | — | — | |
| Yuan et al.Supervision=Strong2017.03 | 18.8 | 26.1 | 33.6 | 42.6 | 51.4 | — | — | — | |
| Yuan et al.2019.06 | 18.8 | 26.1 | 33.6 | 42.6 | 51.4 | — | — | — | |
| Yuan et al.2017.03 | 18.8 | 26.1 | 33.6 | 42.6 | 51.4 | — | — | — | |
| Yuan et al.2019.11 | 17.8 | 27.8 | 36.5 | — | — | — | — | — | |
| RNN-based Action Detection Agentconfiguration=full, backbone=VGG-162015.11 | 17.1 | 26.4 | 36 | 44 | 48.9 | — | — | — | |
| Yeung et. al.2017.03 | 17.1 | 26.4 | 36 | 44 | 48.9 | — | — | — | |
| Yeung et al.Supervision=Strong2017.03 | 17.1 | 26.4 | 36 | 44 | 48.9 | — | — | — | |
| Yeung et al.2019.06 | 17.1 | 26.4 | 36 | 44 | 48.9 | — | — | — | |
| Yeung et al.2019.11 | 17.1 | 26.4 | 36 | — | — | — | — | — | |
| Yeung et al.2017.03 | 17.1 | 26.4 | 36 | 44 | 48.9 | — | — | — | |
| Richard et. al.2017.03 | 15.2 | 23.2 | 30 | 35.7 | 39.7 | — | — | — | |
| Richard et al.Supervision=Strong2017.03 | 15.2 | 23.2 | 30 | 35.7 | 39.7 | — | — | — | |
| Richard et al.2019.06 | 15.2 | 23.2 | 30 | 35.7 | 39.7 | — | — | — | |
| Richard et al.2017.03 | 15.2 | 23.2 | 30 | 35.7 | 39.7 | — | — | — | |
| Oneata et al.features=dense trajectories with video-level CNN classification2015.11 | 14.4 | 20.8 | 27 | 33.6 | 36.6 | — | — | — | |
| Oneata et. al.2017.03 | 14.4 | 20.8 | 27 | 33.6 | 36.6 | — | — | — | |
| Oneata et al.Supervision=Strong2017.03 | 14.4 | 20.8 | 27 | 33.6 | 36.6 | — | — | — | |
| Oneata et al.2019.06 | 14.4 | 20.8 | 27 | 33.6 | 36.6 | — | — | — | |
| Oneata et al.2017.03 | 14.4 | 20.8 | 27 | 33.6 | 36.6 | — | — | — | |
| Escorcia et al.2019.06 | 13.9 | — | — | — | — | — | — | — | |
| Escorcia et al.2017.03 | 13.9 | — | — | — | — | — | — | — | |
| UntrimmedNetSupervision=Weak, selection_module=soft2017.03 | 13.7 | 21.1 | 28.2 | 37.7 | 44.4 | — | — | — | |
| Heilbron et al.2019.06 | 13.5 | — | — | — | — | — | — | — | |
| Heilbron et al.2017.03 | 13.5 | — | — | — | — | — | — | — | |
| RNN-based Action Detection Agentconfiguration=w/o d_pred2015.11 | 12.4 | 19.3 | 26 | 32.5 | 37 | — | — | — | |
| RNN-based Action Detection Agentconfiguration=w/o d_obs2015.11 | 9.3 | 15.2 | 20.6 | 26.5 | 31.2 | — | — | — | |
| RNN-based Action Detection Agentconfiguration=w/o d_obs w/o d_pred2015.11 | 8.6 | 14.6 | 20 | 27.1 | 33.3 | — | — | — | |
| Wang et al.features=combined dense trajectories and CNN features2015.11 | 8.3 | 11.7 | 14 | 17 | 18.2 | — | — | — | |
| Wang et. al.2017.03 | 8.3 | 11.7 | 14 | 17 | 18.2 | — | — | — | |
| Wang et al.2019.06 | 8.3 | 11.7 | 14 | 17 | 18.2 | — | — | — | |
| Wang et al.2017.03 | 8.3 | 11.7 | 14 | 17 | 18.2 | — | — | — | |
| CNN with NMS2015.11 | 6.4 | 9.6 | 12.8 | 16.7 | 18.5 | — | — | — | |
| LSTM with NMS2015.11 | 5.6 | 7.8 | 10.3 | 13.9 | 15.7 | — | — | — | |
| RNN-based Action Detection Agentconfiguration=w/o loc2015.11 | 5.5 | 9.9 | 16.2 | 22.7 | 27.5 | — | — | — | |
| Karaman et al.features=dense trajectories only2015.11 | 0.9 | 1.4 | 2.1 | 3.4 | 4.6 | — | — | — | |
| Karaman et al.2019.06 | 0.9 | 1.4 | 2.1 | 3.4 | 4.6 | — | — | — | |
| Karaman et al.2017.03 | 0.9 | 1.4 | 2.1 | 3.4 | 4.6 | — | — | — | |
| BMNclassifier=UNet2019.07 | 0.388 | 47.4 | 56 | — | — | — | 29.7 | 20.5 | |
| BSNclassifier=UNet2019.07 | 0.369 | 45 | 53.5 | — | — | — | 28.4 | 20 | |
| BMNclassifier=SCNN-cls2019.07 | 0.322 | 40.2 | 45.7 | — | — | — | 24.5 | 17 | |
| BSNclassifier=SCNN-cls2019.07 | 0.294 | 36.6 | 43.1 | — | — | — | 22.4 | 15 | |
| TURNclassifier=SCNN-cls2019.07 | 0.256 | 33.2 | 44.1 | — | — | — | 14.6 | 7.7 | |
| TURNclassifier=UNet2019.07 | 0.245 | 35.3 | 46.3 | — | — | — | 14.1 | 6.3 | |
| SSTclassifier=SCNN-cls2019.07 | 0.23 | — | — | — | — | — | — | — | |
| SSTclassifier=UNet2019.07 | 0.2 | 31.5 | 41.2 | — | — | — | 10.9 | 4.7 | |
| C3D + LinearInterp2018.11 | — | — | — | — | — | 0.37 | — | — | |
| CDC2018.11 | — | — | — | — | — | 0.444 | — | — | |
| Conv & De-conv2018.11 | — | — | — | — | — | 0.417 | — | — | |
| LSTMFeatures=two-stream2017.07 | — | — | — | — | — | 0.393 | — | — | |
| LSTM2018.11 | — | — | — | — | — | 0.393 | — | — | |
| MultiLSTMFeatures=two-stream2017.07 | — | — | — | — | — | 0.413 | — | — | |
| MultiLSTM2018.11 | — | — | — | — | — | 0.413 | — | — | |
| Predictive-corrective2018.11 | — | — | — | — | — | 0.389 | — | — | |
| REDFeatures=two-stream2017.07 | — | — | — | — | — | 0.453 | — | — | |
| RED2018.11 | — | — | — | — | — | 0.453 | — | — | |
| Single-frame CNN2018.11 | — | — | — | — | — | 0.347 | — | — | |
| TRN2018.11 | — | — | — | — | — | 0.472 | — | — | |
| two-streamFeatures=two-stream2017.07 | — | — | — | — | — | 0.362 | — | — | |
| Two-stream CNN2018.11 | — | — | — | — | — | 0.362 | — | — |