Action Recognition on Something-Something V1
70Top-1 AccInternVideo
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InternVideo2022.12 | 70 | — | |
| VideoMAE V2-gBackbone=ViT-g2023.03 | 68.7 | 91.9 | |
| VideoMAE V2-HBackbone=ViT-H2023.03 | 66.6 | 90.8 | |
| UniFormer2023.03 | 61 | 87.6 | |
| VideoMAE2022.12 | 60.9 | — | |
| TDNBackbone=ResNet101, Frames=8+16, GFLOPS=1982020.12 | 56.8 | 84.1 | |
| TDN2023.03 | 56.8 | 84.1 | |
| RSANet-R50ENpre-trained=IN, #frame=8+16, FLOPS x clips=108 Gx22021.11 | 56.1 | 82.8 | |
| RSANet-R50ENpre-trained=IN, #frame=8+16, FLOPS x clips=108 Gx12021.11 | 55.5 | 82.6 | |
| TDNBackbone=ResNet101, Frames=16, GFLOPS=1322020.12 | 55.3 | 83.3 | |
| GSMBackbone=Inception V3, Frames=fusion, GFLOPS=2682020.12 | 55.2 | — | |
| GSM2023.03 | 55.2 | — | |
| TDNBackbone=ResNet50, Frames=8+16, GFLOPS=1082020.12 | 55.1 | 82.9 | |
| RNL TSMEnBackbone=R101 + R50, Frame x Clip x Crop=144=(8+16)x2x32020.07 | 54.1 | 82.2 | |
| TDNBackbone=ResNet101, Frames=8, GFLOPS=662020.12 | 54.1 | 81.9 | |
| RSANet-R50pre-trained=IN, #frame=16, FLOPS x clips=72 Gx12021.11 | 54 | 81.1 | |
| TDNBackbone=ResNet50, Frames=16, GFLOPS=722020.12 | 53.9 | 82.1 | |
| spatial-temporal discriminative filter banksBackbone=ResNet152, Pre-trained=ImageNet2019.08 | 53.4 | 81.8 | |
| CorrNet-101Pretrain=Sports1M, Two stream=False2019.06 | 53.3 | — | |
| CorrNetBackbone=ResNet101, Frames=32x30, GFLOPS=224x30, Pre-trained=Sports1M2020.12 | 53.3 | — | |
| CorrNet2023.03 | 53.3 | — | |
| MARS+RGB+FlowPretrain=Kinetics, Two stream=True2019.06 | 53 | — | |
| RNL TSMEnBackbone=ResNet-50, Frame x Clip x Crop=144=(8+16)x2x32020.07 | 52.7 | 81.5 | |
| TEINetBackbone=ResNet50, Frames=8+16, GFLOPS=992020.12 | 52.5 | — | |
| CT-Net-R50pre-trained=IN, #frame=16, FLOPS x clips=75 Gx12021.11 | 52.5 | 80.9 | |
| TEINet2023.03 | 52.5 | — | |
| TDNBackbone=ResNet50, Frames=8, GFLOPS=362020.12 | 52.3 | 80.6 | |
| TEApre-trained=IN, #frame=16, FLOPS x clips=70 Gx32021.11 | 52.3 | 81.9 | |
| MSNet-TSM-R50pre-trained=IN, #frame=16, FLOPS x clips=67 Gx12021.11 | 52.1 | 82.3 | |
| TEABackbone=ResNet50, Frames=16, GFLOPS=702020.12 | 51.9 | 80.3 | |
| RSANet-R50pre-trained=IN, #frame=8, FLOPS x clips=36 Gx12021.11 | 51.9 | 79.6 | |
| TEA2023.03 | 51.9 | 80.3 | |
| MARS+RGBPretrain=Kinetics, Two stream=False2019.06 | 51.7 | — | |
| CorrNet-101Pretrain=None, Two stream=False, Note=Multi-clip sampling2019.06 | 51.7 | — | |
| CorrNetBackbone=ResNet101, Frames=32x30, GFLOPS=224x302020.12 | 51.7 | — | |
| SmallBigEnBackbone=ResNet-50, Frame x Clip x Crop=144=(8+16)x2x32020.07 | 51.4 | 80.7 | |
| RNL TSMEnBackbone=ResNet-50, Frame x Clip x Crop=24=(8+16)x1x12020.07 | 51.3 | 80.6 | |
| RNL TSMBackbone=ResNet-50, Frame x Clip x Crop=96=16x2x32020.07 | 51 | 80.3 | |
| CorrNet-101Pretrain=None, Two stream=False2019.06 | 50.9 | — | |
| MSNet-R50Backbone=TSM-ResNet50, #3D=false, #Optical flow=false, #Frames=82021.03 | 50.9 | — | |
| CorrNet-R101#frame=32, FLOPS x clips=187 Gx102021.11 | 50.9 | — | |
| RNL TSMBackbone=ResNet-101, Frame x Clip x Crop=48=8x2x32020.07 | 50.8 | 79.8 | |
| STMBackbone=ResNet50, Frames=16x30, GFLOPS=67x302020.12 | 50.7 | 80.4 | |
| STM-R50pre-trained=IN, #frame=16, FLOPS x clips=67 Gx302021.11 | 50.7 | 80.4 | |
| TANetBackbone=ResNet50, Frames=8+16, GFLOPS=992020.12 | 50.6 | 79.3 | |
| TANet2023.03 | 50.6 | 79.3 | |
| V4DBackbone=ResNet50, Frames=8x4, GFLOPS=167.62020.12 | 50.4 | — | |
| SmallBigNetBackbone=ResNet50, Frames=8+16, GFLOPS=1572020.12 | 50.4 | 80.5 | |
| V4D2023.03 | 50.4 | — | |
| spatial-temporal discriminative filter banksBackbone=ResNet50, Pre-trained=ImageNet2019.08 | 50.1 | 79.5 | |
| DFB-NetBackbone=3D ResNet50, #3D=true, #Optical flow=false, #Frames=162021.03 | 50.1 | — | |
| AFNet-TSM (RT=0.8)Dynamic=true, Backbone=AF-ResNet101, Frames=12, GFLOPS=48.92022.11 | 50.1 | — | |
| SmallBigBackbone=ResNet-50, Frame x Clip x Crop=96=16x2x32020.07 | 50 | 79.8 | |
| AFNet-TSM (RT=0.8)Dynamic=true, Backbone=AF-ResNet50, Frames=12, GFLOPS=31.72022.11 | 49.9 | — | |
| AFNet-TSM (RT=0.4)Dynamic=true, Backbone=AF-ResNet101, Frames=12, GFLOPS=42.12022.11 | 49.8 | — | |
| TRGBackbone=Inception-V3, # Frames=32, # Params=28.3M, FLOPS=187.3G2019.08 | 49.7 | 85.3 | |
| TSMEnBackbone=ResNet-50, Frame x Clip x Crop=24=(8+16)x1x12020.07 | 49.7 | 78.5 | |
| TSMBackbone=ResNet50, Frames=8+16, GFLOPS=982020.12 | 49.7 | 78.5 | |
| TSM2023.03 | 49.7 | 78.5 | |
| TRGBackbone=ResNet-50, # Frames=32, # Params=28.4M, FLOPS=132.2G2019.08 | 49.5 | 86.1 | |
| RNL TSMBackbone=ResNet-50, Frame x Clip x Crop=48=8x2x32020.07 | 49.5 | 78.4 | |
| ECO-En Lite-RGB+FlowBackbone=BNIncep+3D Res18, #3D=true, #Optical flow=true, #Frames=92+922021.03 | 49.5 | — | |
| Ours-cumulativeBackbone=TSM-ResNet50, #3D=false, #Optical flow=false, #Frames=all, #Clusters=82021.03 | 49.5 | — | |
| RNL TSMBackbone=ResNet-50, Frame x Clip x Crop=16=16x1x12020.07 | 49.4 | — | |
| CorrNet-50Pretrain=None, Two stream=False2019.06 | 49.3 | — | |
| CorrNetBackbone=ResNet50, Frames=32x10, GFLOPS=115x102020.12 | 49.3 | — | |
| ir-CSN-152#frame=32, FLOPS x clips=97 Gx102021.11 | 49.3 | — | |
| STMBackbone=ResNet50, #3D=false, #Optical flow=false, #Frames=82021.03 | 49.2 | — | |
| AFNet-TSM (RT=0.4)Dynamic=true, Backbone=AF-ResNet50, Frames=12, GFLOPS=27.92022.11 | 49 | — | |
| TEADynamic=false, Backbone=ResNet50, Frames=8, GFLOPS=352022.11 | 48.9 | — | |
| GST-50Pretrain=ImageNet, Two stream=False2019.06 | 48.6 | — | |
| GSTBackbone=ResNet50, Frames=16, GFLOPS=592020.12 | 48.6 | 77.9 | |
| TAMBackbone=bLResNet50, Frames=16x2, GFLOPS=47.72020.12 | 48.4 | 78.8 | |
| SmallBigBackbone=ResNet-50, Frame x Clip x Crop=48=8x2x32020.07 | 48.3 | 78.1 | |
| S3D-GPretrain=ImageNet, Two stream=True2019.06 | 48.2 | — | |
| S3D-GBackbone=Inception, #3D=true, #Optical flow=false, #Frames=642021.03 | 48.2 | — | |
| S3D-GBackbone=Inception, Frames=64, GFLOPS=71.382020.12 | 48.2 | 78.7 | |
| TRGBackbone=ResNet-50, # Frames=16, # Params=28.4M, FLOPS=66.4G2019.08 | 48.1 | 80.4 | |
| AdaFocus-TSMDynamic=true, Backbone=MobileNetV2+ResNet50, Frames=8+12, GFLOPS=33.72022.11 | 48.1 | — | |
| TSM + CGCBackbone=ResNet-50, Param=24.09M, Frame=162019.10 | 47.87 | 77.22 | |
| TrajectoryNetBackbone=ResNet18, Pre-trained=Kinetics-4002019.08 | 47.8 | — | |
| TrajectoryNetBackbone=S3DResNet-18, # Frames=162019.08 | 47.8 | — | |
| bLVNet-TAMDynamic=false, Backbone=bLResNet50, Frames=16, GFLOPS=35.12022.11 | 47.8 | — | |
| TRGBackbone=Inception, # Frames=32, # Params=14.1M, FLOPS=64.5G2019.08 | 47.5 | 80.2 | |
| RNL TSMBackbone=ResNet-50, Frame x Clip x Crop=8=8x1x12020.07 | 47.3 | — | |
| TANetDynamic=false, Backbone=ResNet50, Frames=8, GFLOPS=332022.11 | 47.3 | — | |
| TRGBackbone=Inception-V3, # Frames=16, # Params=28.3M, FLOPS=94.6G2019.08 | 47.2 | 78.9 | |
| TSMBackbone=ResNet-50, Frame x Clip x Crop=16=16x1x12020.07 | 47.2 | 77.1 | |
| TSM-R50pre-trained=IN, #frame=16, FLOPS x clips=65 Gx12021.11 | 47.2 | 77.1 | |
| SmallBigDynamic=false, Backbone=ResNet50, Frames=8, GFLOPS=522022.11 | 47 | — | |
| AdaFuse-TSMDynamic=true, Backbone=ResNet50, Frames=8, GFLOPS=31.52022.11 | 46.8 | — | |
| P3D + CGC 1D & 3DBackbone=ResNet-50, Param=25.62M, Frame=32 x 302019.10 | 46.73 | 76.04 | |
| Ours-slopeBackbone=TSM-ResNet50, #3D=false, #Optical flow=false, #Frames=all, #Clusters=82021.03 | 46.7 | — | |
| TSMBackbone=ResNet-50, Param=23.86M, Frame=162019.10 | 46.61 | 76.18 | |
| ECO liteBackbone=BN-Inception+ResNet18, Pre-trained=Kinetics-4002019.08 | 46.4 | — | |
| ECOenBackbone=Inception+3DResNet-18, # Frames=92, # Params=150.0M, FLOPS=267G2019.08 | 46.4 | — | |
| ECO En LiteBackbone=BNInc+Res18, Param=150M, Frame=922019.10 | 46.4 | — | |
| ECO-En LiteBackbone=BNIncep+3D Res18, #3D=true, #Optical flow=false, #Frames=922021.03 | 46.4 | — | |
| ECO_En LiteBackbone=BNIncep+R18, Frames=92, GFLOPS=2672020.12 | 46.4 | — | |
| P3D + CGC 3DBackbone=ResNet-50, Param=25.61M, Frame=32 x 302019.10 | 46.35 | 75.97 |