Self Supervised Action Recognition On Ucf101

评估指标

3-fold Accuracy

Frozen

Pre-Training Dataset

评测结果

各个模型在此基准测试上的表现结果

				Paper Title	Repository
VideoMAE V2-g	99.6	-	-	VideoMAE V2: Scaling Video Masked Autoencoders with Dual Masking
MVD (ViT-B)	97.5	false	Kinetics400	Masked Video Distillation: Rethinking Masked Feature Modeling for Self-supervised Video Representation Learning
SSL-KD (R21D-18)	97.3	false	Kinetics400	A Large-Scale Analysis on Self-Supervised Video Representation Learning	-
M3Video	96.5	false	Kinetics400	Masked Motion Encoding for Self-Supervised Video Representation Learning
pBYOL	96.3	false	Kinetics400	A Large-Scale Study on Unsupervised Spatiotemporal Representation Learning
VideoMAE	96.1	false	Kinetics400	VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training
SCE (R3D-50)	95.3	false	Kinetics400	Similarity Contrastive Estimation for Image and Video Soft Contrastive Self-Supervised Learning
MMV TSM-50x2	95.2	false	Audioset + Howto100M	Self-Supervised MultiModal Versatile Networks
XKD (ViT-B/112/16)	94.1	-	Kinetics400	XKD: Cross-modal Knowledge Distillation with Domain Alignment for Video Representation Learning
CVRL (R3D-152 2x; K600)	93.9	false	Kinetics600	Spatiotemporal Contrastive Video Representation Learning
RSPNet	93.7	false	Kinetics400	RSPNet: Relative Speed Perception for Unsupervised Video Representation Learning
XKD-Modality-Agnostic (ViT-B/112/16)	93.4	-	-	XKD: Cross-modal Knowledge Distillation with Domain Alignment for Video Representation Learning
CVRL (R3D-50; K600)	93.4	false	Kinetics600	Spatiotemporal Contrastive Video Representation Learning
VideoMS (ViT-B)	93.4	false	no extra data	EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens
BraVe:V-FA (TSM-50x2)	93.1	false	-	Broaden Your Views for Self-Supervised Video Learning
CrissCross (AudioSet)	92.4	false	AudioSet	Self-Supervised Audio-Visual Representation Learning with Relaxed Cross-Modal Synchronicity
CVRL (R3D-50; K400)	92.2	false	Kinetics400	Spatiotemporal Contrastive Video Representation Learning
AVID+CMA (Modified R2+1D-18 on Audioset)	91.5	false	Audioset (Audio+Video)	Audio-Visual Instance Discrimination with Cross-Modal Agreement
CrissCross (Kinetics400)	91.5	false	Kinetics400	Self-Supervised Audio-Visual Representation Learning with Relaxed Cross-Modal Synchronicity
VideoMAE(no extra data)	91.3	false	no extra data	VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training

0 of 53 row(s) selected.