Action Recognition on Kinetics
82.8Top-1 AccR(2+1)D-152 + SC-Sampler
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| R(2+1)D-152 + SC-Samplerpre-training=IG-Kinetics, Input type=RGB, short edge scaling=true2019.05 | 82.8 | 95.3 | — | |
| R(2+1)D-152pre-training=IG-Kinetics, Input type=RGB, short edge scaling=true2019.05 | 81.3 | 95.1 | — | |
| R(2+1)D-152pre-training=IG-Kinetics, Input type=RGB2019.05 | 80.5 | 94.6 | — | |
| R(2+1)D-34pre-training=IG-Kinetics, Input type=RGB, short edge scaling=true2019.05 | 79.6 | 94.2 | — | |
| R(2+1)D-34pre-training=IG-Kinetics, Input type=RGB2019.05 | 79.1 | 93.9 | — | |
| 3-stream SATTpre-training=ImageNet, Input type=RGB + flow + audio2019.05 | 77.7 | 93.2 | — | |
| NL I3Dpre-training=ImageNet, Input type=RGB2019.05 | 77.7 | 93.3 | — | |
| NL I3D-101ImageNet pre-train=true, GFLOPs=359, clips=302019.06 | 77.7 | — | — | |
| NL I3D-50ImageNet pre-train=true, GFLOPs=282, clips=302019.06 | 76.5 | — | — | |
| MSNet-R50flow=false, #frame=16, Backbone FLOPs=67G, #clips=10, speed (V/s)=31.22020.07 | 76.4 | — | — | |
| TSM + Non-local + CGCBackbone=ResNet-50, Param=31.43M2019.10 | 76.4 | 92.5 | — | |
| TSM + CGCBackbone=ResNet-50, Param=24.07M2019.10 | 76.06 | 92.5 | — | |
| I3D-Two-Streampre-training=ImageNet, Input type=RGB + flow2019.05 | 75.7 | 92 | — | |
| TSM + Non-localBackbone=ResNet-50, Param=31.22M2019.10 | 75.6 | 92.15 | — | |
| Rep-flow (R(2+1)D)flow=false, #frame=32, Backbone FLOPs=152G, #clips=25, speed (V/s)=2.02020.07 | 75.5 | — | — | |
| R(2+1)D-Two-Streampre-training=Sports-1M, Input type=RGB + flow2019.05 | 75.4 | 91.9 | — | |
| FASTER32ImageNet pre-train=false, GFLOPs=67.7, clips=82019.06 | 75.3 | — | — | |
| MSNet-R50flow=false, #frame=8, Backbone FLOPs=34G, #clips=10, speed (V/s)=54.22020.07 | 75 | — | — | |
| S3D-GImageNet pre-train=true, GFLOPs=71.42019.06 | 74.7 | — | — | |
| A^2-NetImageNet pre-train=true, GFLOPs=40.8, clips=302019.06 | 74.6 | — | — | |
| TSMBackbone=ResNet-50, Param=23.86M2019.10 | 74.12 | 91.21 | — | |
| R(2+1)D Two-streamflow=true, #frame=32+32, Backbone FLOPs=304G, #clips=115, speed (V/s)=0.22020.07 | 73.9 | — | — | |
| STMflow=false, #frame=16, Backbone FLOPs=67G, #clips=302020.07 | 73.7 | — | — | |
| TSM (reproduced)flow=false, #frame=8, Backbone FLOPs=33G, #clips=10, speed (V/s)=64.12020.07 | 73.5 | — | — | |
| MF-Net#Params=8.0 M, FLOPs=11.1 G2018.07 | 72.8 | 90.4 | — | |
| MF-NetImageNet pre-train=true, GFLOPs=11.1, clips=502019.06 | 72.8 | — | — | |
| S3DImageNet pre-train=false, GFLOPs=66.42019.06 | 72.2 | — | — | |
| I3DImageNet pre-train=false, GFLOPs=108, clips=42019.06 | 72.1 | — | — | |
| R(2+1)D-RGB#Params=63.6 M, FLOPs=152.4 G2018.07 | 72 | 90 | — | |
| R(2+1)D-34ImageNet pre-train=false, GFLOPs=152, clips=1152019.06 | 72 | — | — | |
| R(2+1)Dflow=false, #frame=32, Backbone FLOPs=152G, #clips=115, speed (V/s)=8.72020.07 | 72 | — | — | |
| R(2+1)D-34pre-training=Sports-1M, Input type=RGB2019.05 | 71.7 | 90.5 | — | |
| FASTER16ImageNet pre-train=false, GFLOPs=14.4, clips=162019.06 | 71.7 | — | — | |
| I3D-RGB#Params=12.1 M, FLOPs=107.9 G2018.07 | 71.1 | 89.3 | — | |
| ECOImageNet pre-train=false2019.06 | 70 | — | — | |
| S3D#Params=8.8 M, FLOPs=66.4 G2018.07 | 69.4 | 89.1 | — | |
| S3DImageNet pre-train=false, GFLOPs=66.42019.06 | 69.4 | — | — | |
| ARTNetImageNet pre-train=false, GFLOPs=23.5, clips=2502019.06 | 69.2 | — | — | |
| STCImageNet pre-train=false2019.06 | 68.7 | — | — | |
| Rep-flow (ResNet-50)flow=false, #frame=32, Backbone FLOPs=132G, #clips=25, speed (V/s)=3.72020.07 | 68.5 | — | — | |
| I3DImageNet pre-train=false, GFLOPs=108, clips=42019.06 | 68.4 | — | — | |
| ResNet-50 Two-streamflow=true, #frame=32+32, Backbone FLOPs=264G, #clips=25, speed (V/s)=0.22020.07 | 64.5 | — | — | |
| ConvNet+LSTM#Params=9 M2018.07 | 63.3 | — | — | |
| Two-Stream#Params=12 M2018.07 | 62.2 | — | — | |
| ResNet-50flow=false, #frame=32, Backbone FLOPs=132G, #clips=25, speed (V/s)=22.82020.07 | 61.3 | — | — | |
| RGBType=Frame-based, Modality=RGB2018.01 | 57 | 77.3 | — | |
| RVM-BSize(M)=1172025.12 | 53.1 | — | — | |
| RVM-SSize(M)=342025.12 | 49.6 | — | — | |
| Optical FlowType=Frame-based, Modality=Optical Flow2018.01 | 49.5 | 71.9 | — | |
| 4DS-BSize(M)=91, config=dist from e2025.12 | 46.4 | — | — | |
| VideoMAE-BSize(M)=872025.12 | 38.9 | — | — | |
| MS-G3DBones=true2020.08 | 38 | 60.9 | — | |
| ST-TR-agcnBones=true2020.08 | 38 | 60.5 | — | |
| Sym-GNNmultitasking_prediction=true2019.10 | 37.2 | 58.1 | — | |
| ST-TRBones=true2020.08 | 37 | 59.7 | — | |
| DGNN2019.10 | 36.9 | 56.9 | — | |
| DGNNBones=true2020.08 | 36.9 | 59.6 | — | |
| Sym-GNNmultitasking_prediction=false2019.10 | 36.4 | 57.4 | — | |
| 2s-AGCN2019.10 | 36.1 | 58.7 | — | |
| ST-TR-agcnBones=false2020.08 | 36.1 | 58.7 | — | |
| 2s-AGCNBones=true2020.08 | 36.1 | 58.7 | — | |
| 4DS-BSize(M)=912025.12 | 35.7 | — | — | |
| S-TRBones=true2020.08 | 35.4 | 57.9 | — | |
| SANBones=false2020.08 | 35.1 | 55.7 | — | |
| AS-GCN2019.10 | 34.8 | 56.3 | — | |
| AS-GCNBones=false2020.08 | 34.8 | 56.5 | — | |
| T-TR-agcnBones=true2020.08 | 34.7 | 56.4 | — | |
| ST-TRBones=false2020.08 | 34.5 | 57.6 | — | |
| T-TR-agcnBones=false2020.08 | 34.4 | 57.1 | — | |
| STGR-GCN2019.10 | 33.6 | 56.1 | — | |
| T-TRBones=true2020.08 | 33.1 | 55.86 | — | |
| S-TRBones=false2020.08 | 32.4 | 55.3 | — | |
| T-TRBones=false2020.08 | 32.4 | 55.2 | — | |
| ST-GCNType=Skeleton-based2018.01 | 30.7 | 52.8 | — | |
| ST-GCN2019.10 | 30.7 | 52.8 | — | |
| ST-GCNBones=false2020.08 | 30.7 | 52.8 | — | |
| 4DS-SSize(M)=242025.12 | 28.1 | — | — | |
| Temporal Conv.Type=Skeleton-based2018.01 | 20.3 | 40 | — | |
| Temporal Conv2019.10 | 20.3 | 40 | — | |
| Deep LSTMType=Skeleton-based2018.01 | 16.4 | 35.3 | — | |
| Deep LSTM2019.10 | 16.4 | 35.3 | — | |
| Feature Enc.Type=Skeleton-based2018.01 | 14.9 | 25.8 | — | |
| Feature Encoding2019.10 | 14.9 | 25.8 | — | |
| AGCNSkeleton Source=2D skeletons (OpenPose)2021.04 | — | — | 36.1 | |
| AS-GCNSkeleton Source=2D skeletons (OpenPose)2021.04 | — | — | 34.8 | |
| CapFSARPre-training=BLIP-ViT-B, Shot=1-shot2026.03 | — | — | 83.5 | |
| CapFSARPre-training=BLIP-ViT-B, Shot=5-shot2026.03 | — | — | 92.2 | |
| CLIP-CPM2CPre-training=CLIP-ViT-B, Shot=1-shot2026.03 | — | — | 91 | |
| CLIP-CPM2CPre-training=CLIP-ViT-B, Shot=5-shot2026.03 | — | — | 95.5 | |
| CLIP-FSARPre-training=CLIP-ViT-B, Shot=1-shot2026.03 | — | — | 94.8 | |
| CLIP-FSARPre-training=CLIP-ViT-B, Shot=5-shot2026.03 | — | — | 95.4 | |
| DGNNSkeleton Source=2D skeletons (OpenPose)2021.04 | — | — | 36.9 | |
| EMP-NetPre-training=CLIP-ViT-B, Shot=1-shot2026.03 | — | — | 89.1 | |
| EMP-NetPre-training=CLIP-ViT-B, Shot=5-shot2026.03 | — | — | 93.5 | |
| FSAR-LLaVAPre-training=Video-LLaVA-7B, Knowledge Prompt=Unknown, Shot=1-shot2026.03 | — | — | 94.8 | |
| FSAR-LLaVAPre-training=Video-LLaVA-7B, Knowledge Prompt=Unknown, Shot=5-shot2026.03 | — | — | 97.4 | |
| FSAR-LLaVAPre-training=Video-LLaVA-7B, Knowledge Prompt=Known, Shot=1-shot2026.03 | — | — | 95.3 | |
| FSAR-LLaVAPre-training=Video-LLaVA-7B, Knowledge Prompt=Known, Shot=5-shot2026.03 | — | — | 98.1 | |
| FSAR-Qwen2Pre-training=Qwen2-VL-7B, Knowledge Prompt=Unknown, Shot=1-shot2026.03 | — | — | 93.4 | |
| FSAR-Qwen2Pre-training=Qwen2-VL-7B, Knowledge Prompt=Unknown, Shot=5-shot2026.03 | — | — | 97.1 |