Action Recognition on Something-Something v1 (test)
65.6Top-1 AccuracyATM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ATMBackbone=ViT-L/14, Pre-training=Merged-2B, Frames × Crops × Clips=16×3×2, GFLOPs=842×62023.07 | 65.6 | — | |
| ATMBackbone=ViT-L/14, Pre-training=WIT-400M, Frames × Crops × Clips=16×3×2, GFLOPs=842×62023.07 | 64 | — | |
| ATMYear=2023, Backbone=ViT-L/14, Pre-training=WIT-400M, Frames × Crops × Clips=16×3×1, GFLOPs=842×32023.07 | 63.7 | — | |
| UniFormerV2Year=2023, Backbone=ViT-L/14, Pre-training=WIT-400M, Frames × Crops × Clips=32×3×1, GFLOPs=52002023.07 | 62.7 | — | |
| ATMBackbone=ViT-B/16, Pre-training=WIT-400M, Frames × Crops × Clips=32×3×1, GFLOPs=378×32023.07 | 61.3 | — | |
| ATM_EnBackbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16+32)×1×1, GFLOPs=4692023.07 | 60 | — | |
| ATM_EnYear=2023, Backbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=2012023.07 | 58.6 | — | |
| ATM_EnBackbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=1112023.07 | 57.6 | — | |
| SIFA-TransformerBackbone=Swin-B, GFLOPS x views=270x32022.06 | 57.3 | 85.1 | |
| ATMBackbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=16×1×1, GFLOPs=1342023.07 | 57.2 | — | |
| BQN EnPretrain=ImageNet + K400, Backbone=TSM R101 + X3D-M, Frames × Crops × Clips=(48+48) × 3 × 2, FLOPs=241G × 3 × 2, #Param.=92M2021.03 | 57.1 | 84.2 | |
| ATMYear=2023, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=32×1×1, GFLOPs=1482023.07 | 57.1 | — | |
| TDN_EnYear=2021, Backbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=1982023.07 | 56.8 | — | |
| CT-Net_ENBackbone=2D (R50)×4, #Frame=8+12+16+24, GFLOPS=2802021.06 | 56.6 | 83.9 | |
| SELFYNet-TSM-R50 EN#frame=8+16, FLOPs x clips=114 G x 22021.02 | 56.6 | 84.4 | |
| ATMBackbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=16×1×1, GFLOPs=742023.07 | 56.3 | — | |
| SIFA-NetBackbone=R101, GFLOPS x views=157x3, Input clip length=642022.06 | 56.1 | 84 | |
| SELFYNet-TSM-R50 EN#frame=8+16, FLOPs x clips=114 G x 12021.02 | 55.8 | 83.9 | |
| SELFY_EnYear=2021, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=1142023.07 | 55.8 | — | |
| SIFA-NetBackbone=R101, GFLOPS x views=78x3, Input clip length=322022.06 | 55.4 | 83.1 | |
| TDNBackbone=R101, GFLOPS x views=1322022.06 | 55.3 | 83.3 | |
| TDNYear=2021, Backbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=16×1×1, GFLOPs=1322023.07 | 55.3 | — | |
| SIFA-NetBackbone=R50, GFLOPS x views=112x3, Input clip length=642022.06 | 55.2 | 83.3 | |
| GSMBackbone=Inception, GFLOPS x views=2682022.06 | 55.2 | — | |
| MSNet-R50 En#frame=16+8, FLOPs=101G x 10, #param=49.2M, flow=false2020.07 | 55.1 | 84 | |
| MSNet-TSM-R50 EN#frame=8+16, FLOPs x clips=101 G x 102021.02 | 55.1 | 84 | |
| TDN_EnYear=2021, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=1082023.07 | 55.1 | — | |
| BQNPretrain=ImageNet, Backbone=TSM R101, Frames × Crops × Clips=48 × 3 × 2, FLOPs=231G × 3 × 2, #Param.=85.4M2021.03 | 54.9 | 81.7 | |
| MSNet-R50 En#frame=16+8, FLOPs=101G x 1, #param=49.2M, flow=false2020.07 | 54.4 | 83.8 | |
| SELFYNet-TSM R50#frame=16, FLOPs x clips=77 G x 12021.02 | 54.3 | 82.9 | |
| BQNPretrain=ImageNet, Backbone=TSM R50, Frames × Crops × Clips=48 × 3 × 2, FLOPs=121G × 3 × 2, #Param.=47.4M2021.03 | 54.3 | 82 | |
| SIFA-NetBackbone=R50, GFLOPS x views=51x3, Input clip length=322022.06 | 54 | 82.2 | |
| MVFNet_EnYear=2021, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×3×2, GFLOPs=99×62023.07 | 54 | — | |
| SSAEnBackbone=ResNet-50, Modality=RGB, Frames=16+82021.05 | 54 | — | |
| TDNBackbone=R50, GFLOPS x views=722022.06 | 53.9 | 82.1 | |
| BQNPretrain=K400, Backbone=X3D-M, Frames × Crops × Clips=48 × 3 × 2, FLOPs=9.7G × 3 × 2, #Param.=6.6M2021.03 | 53.7 | 81.8 | |
| SIFA-NetBackbone=R101, GFLOPS x views=39x3, Input clip length=162022.06 | 53.7 | 82 | |
| CT-NetBackbone=2D R50, #Frame=16×3×2, GFLOPS=4472021.06 | 53.4 | 81.7 | |
| BQNPretrain=ImageNet, Backbone=TSM R50, Frames × Crops × Clips=24 × 3 × 2, FLOPs=60G × 3 × 2, #Param.=47.4M2021.03 | 53.3 | 82 | |
| CorrNetBackbone=R101, GFLOPS x views=224x302022.06 | 53.3 | — | |
| MG-TEABackbone=R1012022.06 | 53.3 | — | |
| MG-TEABackbone=R502022.06 | 53.2 | — | |
| SIFA-NetBackbone=R50, GFLOPS x views=25x3, Input clip length=162022.06 | 52.7 | 81.9 | |
| TSM+W3Backbone=R50, #Frame=16, GFLOPs=67.12020.04 | 52.6 | 81.3 | |
| TSM Two-stream#frame=16+16, FLOPs=129G x 1, #param=48.6M, flow=true, reproduced=true2020.07 | 52.6 | 81.9 | |
| TSM-R50 Two-Streamflow=true, #frame=16+16, FLOPs x clips=129 G x 12021.02 | 52.6 | 81.9 | |
| CT-NetBackbone=2D R50, #Frame=16, GFLOPS=752021.06 | 52.5 | 80.9 | |
| SELFYNet-TSM R50#frame=8, FLOPs x clips=37 G x 12021.02 | 52.5 | 80.8 | |
| TEINetBackbone=R50, GFLOPS x views=992022.06 | 52.5 | — | |
| TEINet_EnYear=2020, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=992023.07 | 52.5 | — | |
| TEA-R50#frame=16, FLOPs x clips=70 G x 302021.02 | 52.3 | 81.9 | |
| TEAPretrain=ImageNet, Backbone=R50, Frames × Crops × Clips=16 × 3 × 10, FLOPs=70G × 3 × 10, #Param.=24.4M2021.03 | 52.3 | 81.9 | |
| TEAYear=2020, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=16×3×10, GFLOPs=70×302023.07 | 52.3 | — | |
| MSNet-R50#frame=16, FLOPs=67G x 1, #param=24.6M, flow=false2020.07 | 52.1 | 82.3 | |
| MSNet-TSM-R50#frame=16, FLOPs x clips=67 G x 12021.02 | 52.1 | 82.3 | |
| MSNetBackbone=R50, GFLOPS x views=672022.06 | 52.1 | 82.3 | |
| MSNetYear=2020, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=16×1×1, GFLOPs=672023.07 | 52.1 | — | |
| PEM+TDLossBackbone=2D R50+TIM, #Frame=16×3×2, GFLOPS=5172021.06 | 52 | — | |
| TEABackbone=2D Res2Net50, #Frame=16, GFLOPS=702021.06 | 51.9 | 80.3 | |
| CT-NetBackbone=2D R50, #Frame=8×3×2, GFLOPS=2242021.06 | 51.7 | 80.1 | |
| BQNPretrain=ImageNet, Backbone=TSM R50, Frames × Crops × Clips=24 × 1 × 1, FLOPs=60G × 1 × 1, #Param.=47.4M2021.03 | 51.7 | 80.5 | |
| CorrNetYear=2020, Backbone=ResNet101, Pre-training=ImageNet-1K, Frames × Crops × Clips=32×3×10, GFLOPs=224×302023.07 | 51.7 | — | |
| Ours-cumulative#Frames=all, #Clusters=162021.03 | 51.4 | — | |
| MUFIPre-train=Multi-Faceted, Evaluation Protocol=fine-tuning2022.01 | 51.2 | 81 | |
| MSNet-R50#frame=8, FLOPs=34G x 1, #param=24.6M, flow=false2020.07 | 50.9 | 80.3 | |
| PEM+TDLossBackbone=2D R50+TIM, #Frame=16, GFLOPS=662021.06 | 50.9 | — | |
| CorrNet-R101#frame=32, FLOPs x clips=187 G x 102021.02 | 50.9 | — | |
| STM#frame=16, FLOPs=67G x 30, #param=24.0M, flow=false2020.07 | 50.7 | 80.4 | |
| STM#Frames=162021.03 | 50.7 | — | |
| TSMBackbone=ResNet-50, Pretrain=Kinetics, Frames=16+16, Modality=RGB+Flow2019.12 | 50.7 | — | |
| SELFYNet-R50#frame=8, FLOPs x clips=37 G x 12021.02 | 50.7 | 79.3 | |
| STM-R50#frame=16, FLOPs x clips=67 G x 302021.02 | 50.7 | 80.4 | |
| STMPre-train=ImageNet2022.01 | 50.7 | 80.4 | |
| STMBackbone=R50, GFLOPS x views=67x302022.06 | 50.7 | 80.4 | |
| TSM 2-streamBackbone=ResNet-50, Modality=RGB+flow, Frames=16+162021.05 | 50.7 | — | |
| BQNPretrain=None, Backbone=X3D-M, Frames × Crops × Clips=48 × 3 × 2, FLOPs=9.7G × 3 × 2, #Param.=6.6M2021.03 | 50.6 | 79.2 | |
| PAN FullPretrain=ImageNet, Backbone=TSM R50, Frames × Crops × Clips=40 × 1 × 2, FLOPs=67.7G × 1 × 22021.03 | 50.5 | 79.2 | |
| PEM+TDLossBackbone=2D R50+TIM, #Frame=8×3×2, GFLOPS=2592021.06 | 50.4 | — | |
| DFB-Net#frame=16, FLOPs=N/A x 1, flow=false2020.07 | 50.1 | 79.5 | |
| CT-NetBackbone=2D R50, #Frame=8, GFLOPS=372021.06 | 50.1 | 78.8 | |
| STDFBPre-train=ImageNet2022.01 | 50.1 | 79.5 | |
| TEINetBackbone=2D R50, #Frame=16, GFLOPS=662021.06 | 49.9 | — | |
| PEM+TDLossBackbone=2D R50+TIM, #Frame=8, GFLOPS=332021.06 | 49.8 | — | |
| ABM-C-inPre-train=ImageNet2022.01 | 49.8 | — | |
| TSMenBackbone=R50, #Frame=8+16, GFLOPs=982020.04 | 49.7 | 78.5 | |
| TSM En#frame=16+8, FLOPs=98G x 1, #param=48.6M, flow=false2020.07 | 49.7 | 78.5 | |
| TSM EnPretrain=ImageNet, Backbone=R50, Frames × Crops × Clips=(16+8) × 1 × 1, FLOPs=98G, #Param.=48.6M2021.03 | 49.7 | 78.5 | |
| TSM_EnYear=2019, Backbone=ResNet50, Pre-training=ImageNet-1K, Frames × Crops × Clips=(8+16)×1×1, GFLOPs=982023.07 | 49.7 | — | |
| ECOen LightRGB+FlowBackbone=BNI+3D R18, #Frame=92+922020.04 | 49.5 | — | |
| ECO Two-stream#frame=92+92, FLOPs=N/A x 1, #param=300M, flow=false2020.07 | 49.5 | — | |
| Ours-cumulative#Frames=all, #Clusters=82021.03 | 49.5 | — | |
| ir-CSNBackbone=3D R152, #Frame=32×1×10, GFLOPS=9672021.06 | 49.3 | — | |
| CorrNetBackbone=3D R50, #Frame=32×1×10, GFLOPS=11502021.06 | 49.3 | — | |
| ir-CSNPretrain=None, Backbone=3D R152, Frames × Crops × Clips=32 × 1 × 10, FLOPs=96.7G × 1 × 102021.03 | 49.3 | — | |
| SmallBigBackbone=R50, GFLOPS x views=1052022.06 | 49.3 | 79.5 | |
| CorrNetBackbone=R50, GFLOPS x views=115x102022.06 | 49.3 | — | |
| STM#Frames=82021.03 | 49.2 | — | |
| I3D+RSTGPre-train=ImageNet+K4002022.01 | 49.2 | 78.8 | |
| TSM+W3Backbone=R50, #Frame=8, GFLOPs=33.52020.04 | 49 | 77.3 | |
| TEABackbone=2D Res2Net50, #Frame=8, GFLOPS=352021.06 | 48.9 | 78.1 |