AI Paper Atlas

Research map / Benchmarks / Kinetics

Kinetics leaderboard

23 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on Kinetics. Best accuracy: 91.2 (STRM, CVPR 2022).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

Kinetics · accuracy

12 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
191.2STRM
ViT
Spatio-Temporal Relation Modeling for Few-Shot Action Recognition
CVPR 2022 · Table 1
Action recognition
282.8R(2+1)D-152 + SC-Sampler; IG-Kinetics; SE
R(2+1)D-152
Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition
CVPR 2019 · Kinetics state-of-the-art comp
Action recognition
377.9Ours ((2+1)D CNN + FcF) + Non-local
(2+1)D CNN
Representation Flow for Action Recognition
CVPR 2019 · state-of-the-art comparison ta
CNN design
476.4MSNet-R50 (16 frames)MotionSqueeze: Neural Motion Feature Learning for Video Understanding
ECCV 2020 · Table 2
Action recognition
575.6SpeedNet
S3D-G
SpeedNet: Learning the Speediness in Videos
CVPR 2020 · SpeedNet performance table
Action recognition
675.4R(2+1)D-Two-Stream
R(2+1)D-34
A Closer Look at Spatiotemporal Convolutions for Action Recognition
CVPR 2018 · Table 5
Action recognition
775.1CorrNet-26
R(2+1)D-26
Video Modeling With Correlation Networks
CVPR 2020 · Table: Correlation networks vs
Action recognition
874.2SINet (K=3)Attend and Interact: Higher-Order Object Interactions for Video Understanding
CVPR 2018 · Table: Pairwise/triplet vs hig
Video-language understanding
972ClusterFit, K=6400
R(2+1)D-34
ClusterFit: Improving Generalization of Visual Representations
CVPR 2020 · Table: Weakly-supervised video
Self-supervised and contrastive
1071.5OursRecognize Actions by Disentangling Components of Dynamics
CVPR 2018 · Table 8
Action recognition
2 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Kinetics · accuracy (val)

12 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
179.4Attention Cluster RGB+Flow+Audio
Inception-ResNet-v2 (RGB/flow), VGG-16 (audio)
Attention Clusters: Purely Attention Based Local Feature Integration for Video Classification
CVPR 2018 · Kinetics results table
Video segmentation and tracking
277.7NL I3D
ResNet-101
Non-Local Neural Networks
CVPR 2018 · Table: Comparisons with state-
CNN features
377.5CoST
ResNet-101, 32 frames
Collaborative Spatiotemporal Feature Learning for Video Action Recognition
CVPR 2019 · Table: Comparison with state-o
Action recognition
477Attention FusionMultimodal Keyless Attention Fusion for Video Classification
AAAI 2018 · Table 3
Video-language understanding
576.4TSM + Non-local + CGC
ResNet-50
Context-Gated Convolution
ECCV 2020 · Table 4
CNN design
676.31DNL
Slow-only R50
Disentangled Non-local Neural Networks
ECCV 2020 · Table 7
CNN features
776GCNet (+all GC)
ResNet-50
Global Context Networks
TPAMI 2020 · Table (Kinetics)
CNN design
875.3FASTER32 (R3D-50)
R3D-50
FASTER Recurrent Networks for Efficient Video Classification
AAAI 2020 · Table: Comparisons to the stat
Action recognition
974.2SINet (dot-product attention)Attend and Interact: Higher-Order Object Interactions for Video Understanding
CVPR 2018 · Table: Prediction accuracy on
Video-language understanding
1073.7Rescale
I3D ResNet50
Is normalization indispensable for training deep neural network?
NeurIPS 2020 · Table 7
CNN features
2 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.