Action Recognition In Videos On Hmdb 51

评估指标

Average accuracy of 3 splits

评测结果

各个模型在此基准测试上的表现结果

模型名称	Average accuracy of 3 splits	Paper Title	Repository
TDD + IDT	65.9	Action Recognition with Trajectory-Pooled Deep-Convolutional Descriptors
TVNet+IDT	72.6	End-to-End Learning of Motion Representation for Video Understanding
R2+1D-BERT	85.10	Late Temporal Modeling in 3D CNN Architectures with BERT for Action Recognition
Multi-stream I3D	80.92	Contextual Action Cues from Camera Sensor for Multi-Stream Action Recognition	-
ADL+ResNet+IDT	74.3	Contrastive Video Representation Learning via Adversarial Perturbations	-
ActionFlowNet	56.4	ActionFlowNet: Learning Motion Representation for Action Recognition	-
ARTNet w/ TSN	70.9	Appearance-and-Relation Networks for Video Classification
VIMPAC	65.9	VIMPAC: Video Pre-Training via Masked Token Prediction and Contrastive Learning
Two-Stream (ImageNet pretrained)	59.4	Two-Stream Convolutional Networks for Action Recognition in Videos
S3D-G (ImageNet, Kinetics-400 pretrained)	75.9	Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification
R[2+1]D-Flow (Kinetics pretrained)	76.4	A Closer Look at Spatiotemporal Convolutions for Action Recognition
R[2+1]D (VideoMoCo)	49.2	VideoMoCo: Contrastive Video Representation Learning with Temporally Adversarial Examples
SO+MaxExp+IDT	85.70	High-order Tensor Pooling with Attention for Action Recognition	-
Flow-I3D (Kinetics pre-training)	77.3	Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset
RepFlow-50 ([2+1]D CNN, FcF, Non-local block)	81.1	Representation Flow for Action Recognition
FASTER32 (Kinetics pretrain)	75.7	FASTER Recurrent Networks for Efficient Video Classification	-
R[2+1]D-RGB (Sports1M pretrained)	66.6	A Closer Look at Spatiotemporal Convolutions for Action Recognition
Dynamic Image Networks + IDT	65.2	Dynamic Image Networks for Action Recognition
Res3D	54.9	ConvNet Architecture Search for Spatiotemporal Feature Learning
Prob-Distill	72.0	Attention Distillation for Learning Video Representations	-

0 of 76 row(s) selected.