Action Classification On Kinetics 400

评估指标

Acc@1

评测结果

各个模型在此基准测试上的表现结果

模型名称	Acc@1	Paper Title	Repository
OmniVec	91.1	OmniVec: Learning robust representations with cross modal sharing	-
X3D-L	77.5	X3D: Expanding Architectures for Efficient Video Recognition
ViT-B-VTN+ ImageNet-21K (84.0 [10])	79.8	Video Transformer Network
MViT-B, 32x3	80.2	Multiscale Vision Transformers
MTV-H (WTS 60M)	89.9	Multiview Transformers for Video Recognition
AdaMAE	81.7	AdaMAE: Adaptive Masking for Efficient Spatiotemporal Learning with Masked Autoencoders
ViC-MAE (ViT-L)	85.1	ViC-MAE: Self-Supervised Representation Learning from Images and Video with Contrastive Masked Autoencoders
MoViNet-A4	80.5	MoViNets: Mobile Video Networks for Efficient Video Recognition
SlowFast 16x8 (ResNet-101)	78.9	SlowFast Networks for Video Recognition
R[2+1]D-RGB (Sports-1M pretrain)	74.3	A Closer Look at Spatiotemporal Convolutions for Action Recognition
X-CLIP(ViT-L/14, CLIP)	87.7	Expanding Language-Image Pretrained Models for General Video Recognition
ip-CSN-152 (IG-65M pretraining)	82.5	Video Classification with Channel-Separated Convolutional Networks
MARS+RGB+Flow (64 frames)	74.9	MARS: Motion-Augmented RGB Stream for Action Recognition	-
TokenLearner 16at18 (L/10)	85.4	TokenLearner: What Can 8 Learned Tokens Do for Images and Videos?
VideoMamba-M800	85.0	VideoMamba: State Space Model for Efficient Video Understanding
TAdaConvNeXt-T	79.1	TAda! Temporally-Adaptive Convolutions for Video Understanding
MAR (50% mask, ViT-B, 16x4)	81.0	MAR: Masked Autoencoders for Efficient Action Recognition
Swin-S (ImageNet-1k pretrain)	80.6	Video Swin Transformer
OMNIVORE (Swin-B)	84.0	Omnivore: A Single Model for Many Visual Modalities
S3D-G (Flow, ImageNet pretrained)	68	Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification

0 of 204 row(s) selected.