List your company
ResearchFormerly WizWand

Action Phase Classification on COIN

54.1Phase AccTimeSformer (Distant Supervision)
28.51635.15841.848.442Nov 17, 2021Nov 28, 2021Dec 10, 2021Dec 22, 2021Jan 2, 2022Jan 14, 2022Jan 26, 2022
Updated 8mo ago

Evaluation results

MethodLinks
TimeSformer (Distant Supervision)Segment Model=TimeSformer, Pretraining Supervision=Unsupervised: distant supervision (ours), Pretraining Dataset=HT100M, Evaluation Protocol=linear probing2022.0154.1
Supervised LearningFraction of Labels=1.02021.1149.18
TimeSformerSegment Model=TimeSformer, Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Evaluation Protocol=linear probing2022.0148.3
VAVAFraction of Labels=1.02021.1147.26
TimeSformerSegment Model=TimeSformer, Pretraining Supervision=Unsupervised: k-means on ASR, Pretraining Dataset=HT100M, Evaluation Protocol=linear probing2022.0146.5
VAVAFraction of Labels=0.52021.1146.18
VAVAFraction of Labels=0.12021.1143.77
Supervised LearningFraction of Labels=0.52021.1140.73
TCCFraction of Labels=1.02021.1140.66
TCNFraction of Labels=1.02021.1140.51
SALFraction of Labels=1.02021.1140.32
LAVFraction of Labels=1.02021.1139.81
TCNFraction of Labels=0.52021.1139.73
TCCFraction of Labels=0.52021.1139.56
VideoCLIPSegment Model=VideoCLIP, Pretraining Supervision=Unsupervised: NCE on ASR, Pretraining Dataset=HT100M, Evaluation Protocol=linear probing2022.0139.4
SALFraction of Labels=0.52021.1139.23
LAVFraction of Labels=0.52021.1138.85
S3DSegment Model=S3D, Pretraining Supervision=Unsupervised: MIL-NCE on ASR, Pretraining Dataset=HT100M, Evaluation Protocol=fine-tuning2022.0137.5
Imagenet FeaturesFraction of Labels=1.02021.1137.43
Supervised LearningFraction of Labels=0.12021.1137.11
LAVFraction of Labels=0.12021.1136.79
TSN (RGB+Flow)Segment Model=TSN (RGB+Flow), Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Evaluation Protocol=fine-tuning2022.0136.5
TCCFraction of Labels=0.12021.1135.87
TCNFraction of Labels=0.12021.1134.87
Imagenet FeaturesFraction of Labels=0.52021.1134.74
SALFraction of Labels=0.12021.1134.69
SlowFastSegment Model=SlowFast, Pretraining Supervision=Supervised: action labels, Pretraining Dataset=Kinetics, Evaluation Protocol=linear probing2022.0132.9
Imagenet FeaturesFraction of Labels=0.12021.1131.32
ClipBERTSegment Model=ClipBERT, Pretraining Supervision=Supervised: captions, Pretraining Dataset=COCO + Visual Genome, Evaluation Protocol=linear probing2022.0130.8
Random FeaturesFraction of Labels=1.02021.1130.38
Random FeaturesFraction of Labels=0.52021.1130.29
Random FeaturesFraction of Labels=0.12021.1129.5