Kinetics Dataset | Papers With Code

The Kinetics dataset is a large-scale, high-quality dataset for human action recognition in videos. The dataset consists of around 500,000 video clips covering 600 human action classes with at least 600 video clips for each action class. Each video clip lasts around 10 seconds and is labeled with a single action class. The videos are collected from YouTube.

Source: Self-supervised Visual Feature Learning with Deep Neural Networks: A Survey

Benchmarks

Add a new result Link an existing benchmark

Task	Dataset Variant	Best Model
Action Classification	Kinetics-400	InternVideo2-6B
Action Classification	Kinetics-600	InternVideo2-6B
Skeleton Based Action Recognition	Kinetics-Skeleton dataset	Structured Keypoint Pooling
Action Classification	Kinetics-700	InternVideo2-6B
Zero-Shot Action Recognition	Kinetics	IMP-MoE-L
Video Prediction	Kinetics-600 12 frames, 64x64	MAGVIT-v2
Spatio-Temporal Action Localization	AVA-Kinetics	VideoMAE V2-g
Few Shot Action Recognition	Kinetics-100	AMeFu-Net
Self-Supervised Action Recognition	Kinetics-600	CVRL
Video Generation	Kinetics-600 12 frames, 64x64	W.A.L.T-L
Self-Supervised Action Recognition	Kinetics-400	CVRL
Action Recognition In Videos	Kinetics-400	CAST-B/16
Text to Video Retrieval	Kinetics-GEB+	FROZEN-revised
Visual Tracking	Kinetics	TAPIR
Action Classification	Kinetics-Sounds	Mirasol3B
Generic Event Boundary Detection	Kinetics-GEBD	FlowGEBD
Video Generation	Kinetics-600 48 frames, 64x64	DVD-GAN
Video Generation	Kinetics-600 12 frames, 128x128	DVD-GAN
Action Classification	MiniKinetics	MARS+RGB+Flow
Video Classification	Kinetics	Multigrid
Action Recognition In Videos	Kinetics-600	Florence
Boundary Detection	Kinetics-400	CASTANET+ Ensemble
Event Segmentation	Kinetics-400	CASTANET+ Ensemble
Text-to-Video Generation	Kinetics	NUWA
Semantic Object Interaction Classification	Kinetics-700	3D ResNet-50
Boundary Captioning	Kinetics-GEB+	ActBERT-revised
Boundary Grounding	Kinetics-GEB+	FROZEN-revised
Skeleton Based Action Recognition	Kinetics-400	STGAT