List your company
ResearchFormerly WizWand

Action Classification on HMDB51

74.5Top-1 AccuracyCross-Ent.
14.80430.30245.861.298Sep 10, 2019Feb 26, 2020Aug 14, 2020Jan 31, 2021Jul 20, 2021Jan 6, 2022Jun 25, 2022
Updated 8mo ago

Evaluation results

MethodLinks
Cross-Ent.Evaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=16 x 112^2, Arch=R(2+1)D-342022.0674.5
COCLREvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=S3D-232022.0662.9
COCLREvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0658.7
TANSettings=finetuned with TAN, Backbone=S3D, Evaluation Protocol=Linear Probing2022.0456.7
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0656.2
Supervised SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=R3D-182022.0656.1
SeCoEvaluation Protocol=Fine-Tuning, Pre-train data (duration)=K400 (28d), Modality=V2020.0855.6
MIL-NCESettings=reproduce of [47], Backbone=S3D, Evaluation Protocol=Linear Probing2022.0455.2
COCLR-RGBEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=S3D-232022.0654.6
MemDPCEvaluation Protocol=Fine-Tuning, Pre-train data (duration)=K400 (28d), Modality=VF2020.0854.5
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=R3D-182022.0654.5
TCLREvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=16x2 x 112^2, Arch=R3D-18, Split Setting=3-split avg2022.0653.6
MIL-NCESettings=reported by [47], Backbone=S3D, Evaluation Protocol=Linear Probing2022.0453.1
TCLREvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16x2 x 112^2, Arch=R3D-18, Split Setting=3-split avg2022.0652.9
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=R3D-18, Split Setting=Split-12022.0652.2
COCLR-RGBEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0652.1
CoCon-EEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=(-) x 224^2, Arch=R3D-18, Input Modality=Optical Flow / Residual, Split Setting=3-split avg2022.0652
CoCon-EEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=(-) x 224^2, Arch=R3D-18, Input Modality=Optical Flow / Residual, Split Setting=3-split avg2022.0652
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=R3D-18, Split Setting=3-split avg2022.0652
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=S3D-23, Split Setting=3-split avg2022.0651.7
Supervised SLICEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=R3D-182022.0650.6
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=32 x 128^2, Arch=S3D-23, Split Setting=Split-12022.0649.4
VTHCLEvaluation Protocol=Fine-Tuning, Pre-train data (duration)=K400 (28d), Modality=V2020.0849.2
SpeedNetEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=16 x 224^2, Arch=S3D-G2022.0648.8
SLICEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=R3D-182022.0648.3
Supervised SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16 x 128^2, Arch=R3D-182022.0646.6
SLICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16 x 128^2, Arch=R3D-182022.0646.2
CoCon-RGBEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=(-) x 224^2, Arch=R3D-18, Split Setting=3-split avg2022.0646
Supervised SLICEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=16 x 128^2, Arch=R3D-182022.0644.9
SLICEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0644.7
VideoMoCoEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=16 x 112^2, Arch=R3D-182022.0643.6
SLICEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=16 x 128^2, Arch=R3D-182022.0641.8
RSPNetEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=16 x 112^2, Arch=R3D-182022.0641.8
MemDPCEvaluation Protocol=Fine-Tuning, Pre-train data (duration)=K400 (28d), Modality=V2020.0841.2
ImageNet PretrainedArchitecture=VGG-M-2048, Number of Parameters=25.4M, Pre-training Dataset=ImageNet2019.0940.5
CoCLREvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0640.2
CoCLR-RGBEvaluation Protocol=Linear Probe, Pre-train Dataset=UCF101, Input Size=32 x 128^2, Arch=S3D-232022.0639.1
CoCon-RGBEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=(-) x 224^2, Arch=R3D-182022.0638.4
IICEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16 x 112^2, Arch=R3D-18, Input Modality=Optical Flow / Residual, Split Setting=3-split avg2022.0638.3
DPC (Ours)Architecture=3D-ResNet34, Number of Parameters=32.6M, Pre-training Dataset=Kinetics-400, Input Resolution=224x2242019.0935.7
DPC (Ours)Architecture=3D-ResNet18, Number of Parameters=14.2M, Pre-training Dataset=Kinetics-400, Input Resolution=128x1282019.0934.5
DPCEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=Kinetics400, Input Size=40 x 128^2, Arch=R3D-182022.0634.5
3D-RotNetArchitecture=3D-ResNet18-full, Number of Parameters=33.6M, Pre-training Dataset=Kinetics-400, Input Resolution=112x1122019.0933.7
3D-ST PuzzleArchitecture=3D-ResNet18-full, Number of Parameters=33.6M, Pre-training Dataset=Kinetics-400, Input Resolution=224x224, Multi-task Learning=Rotation + 3D Puzzle2019.0933.7
VCPEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16 x 112^2, Arch=R3D-18, Split Setting=3-split avg2022.0631.5
MemDPCEvaluation Protocol=Linear Evaluation, Pre-train data (duration)=K400 (28d), Modality=VF2020.0830.5
VCOPEvaluation Protocol=End-to-end Finetuning, Pre-train Dataset=UCF101, Input Size=16 x 112^2, Arch=R3D-18, Split Setting=3-split avg2022.0629.5
OPNArchitecture=VGG-M-2048, Number of Parameters=8.6M, Pre-training Dataset=UCF101/HMDB51, Input Resolution=80x802019.0923.8
None (Rand. Init.)Input Size=16 x 128^2, Arch=R3D-182022.0622.3
Shuffle & LearnArchitecture=CaffeNet, Number of Parameters=58.3M, Pre-training Dataset=UCF101/HMDB51, Input Resolution=227x2272019.0918.1
Random InitializationArchitecture=3D-ResNet18, Number of Parameters=14.2M, Pre-training Dataset=None2019.0917.1