AI Paper Atlas

Research map / Benchmarks / Kinetics-400

Kinetics-400 leaderboard

163 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on Kinetics-400. Best accuracy: 93.6 (Ours, CVPR 2024).

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-400 · accuracy

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

#accuracy ↑MethodPaperTopic
193.6OursOmniVec2 - A Novel Transformer based Network for Large Scale Multimodal and Multitask Learning
CVPR 2024 · Table 2
Cross-modal learning
290.9TubeViT-H
ViT-H
Rethinking Video ViTs: Sparse Video Tubes for Joint Image and Video Learning
CVPR 2023 · Table 1
Video segmentation and tracking
390.22P2S + EVL
EVL
From Isolated Islands to Pangea: Unifying Semantic Space for Human Action Understanding
CVPR 2024 · Table 5
Action recognition
490FluxViT-B
ViT-B
Make Your Training Flexible: Towards Deployment-Efficient Video Models
ICCV 2025 · text
Video-language understanding
590VideoMAE V2-g (64x266^2)
ViT-g
VideoMAE V2: Scaling Video Masked Autoencoders With Dual Masking
CVPR 2023 · Table 5 (K400 table)
Video segmentation and tracking
690UniFormerV2-L 336 (64 frames)
CLIP-ViT-L + K710
UniFormerV2: Unlocking the Potential of Image ViTs for Video Understanding
ICCV 2023 · Table 7
Video segmentation and tracking
789.7EVA
ViT giant
EVA: Exploring the Limits of Masked Visual Representation Learning at Scale
CVPR 2023 · Video action recognition table
Open-set and few-shot detection
889.4ATM ViT-L/14 336 Merged-2B
ViT-L/14
What Can Simple Arithmetic Operations Do for Temporal Modeling?
ICCV 2023 · Table 4
Action recognition
989.1MTV-H (WTS)Multiview Transformers for Video Recognition
CVPR 2022 · Table 2(a)
Action recognition
1088.7ILA-ViT-L/14@336px (16 frames)
CLIP ViT-L/14@336px
Implicit Temporal Modeling with Learnable Alignment for Video Recognition
ICCV 2023 · Table 1
Video segmentation and tracking
118 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-400 · accuracy (val)

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

#accuracy ↑MethodPaperTopic
189.1DejaVid (frozen weights; w-init-one)
VideoMAE V2-g
DejaVid: Encoder-Agnostic Learned Temporal Matching for Video Classification
CVPR 2025 · Table 1
Video segmentation and tracking
288.8MoTED, ViT-L/14, 32 frames
CLIP ViT-L/14
Enhanced Motion-Text Alignment for Image-to-Video Transfer Learning
CVPR 2024 · Table 6
Video-language understanding
387.2ZeroI2V (32x1x3)
CLIP ViT-L/14
ZeroI2V: Zero-Cost Adaptation of Pre-Trained Transformers from Image to Video
ECCV 2024 · Table 4
Video segmentation and tracking
485.12PaPr (z=0.5)
ViT-L MAE
PaPr: Training-Free One-Step Patch Pruning with Lightweight ConvNets for Faster Inference
ECCV 2024 · Table 7
Attention and transformers
584.9STAN-self-B/16
CLIP-B/16
Revisiting Temporal Modeling for CLIP-Based Image-to-Video Knowledge Transferring
CVPR 2023 · Table: Comparison on Kinetics-
Video-language understanding
684.8MAE K400 1600ep
ViT-L
Masked Autoencoders As Spatiotemporal Learners
NeurIPS 2022 · Table 3
Video segmentation and tracking
784.2SIFAR-L-12
Swin-L
Can an Image Classifier Suffice For Action Recognition?
ICLR 2022 · Table: Comparison with Other A
Action recognition
883.6Video-FocalNet-B
ImageNet-1K init
Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition
ICCV 2023 · Table 2
Action recognition
983.6OmniSource
irCSN-152 32x2 (IG-65M pretrain)
Omni-sourced Webly-supervised Learning for Video Recognition
ECCV 2020 · Table 3
Action recognition
1083.5DTF-Transformer
Swin-B
Dynamic Temporal Filtering In Video Models
ECCV 2022 · Table 2
Video segmentation and tracking
29 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-400 · top-5 accuracy

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

#top-5 accuracy ↑MethodPaperTopic
198.9TubeViT-H
ViT-H
Rethinking Video ViTs: Sparse Video Tubes for Joint Image and Video Learning
CVPR 2023 · Table 1
Video segmentation and tracking
298.26P2S + EVL
EVL
From Isolated Islands to Pangea: Unifying Semantic Space for Human Action Understanding
CVPR 2024 · Table 5
Action recognition
396.7Storyboard-guided Alignment
ViT-B/16
Storyboard-guided Alignment for Fine-grained Video Action Recognition
NeurIPS 2025 · Kinetics-400 comparison table
Action recognition
495.5DualFormer-B
IN-21K pretrain
DualFormer: Local-Global Stratified Transformer for Efficient Video Recognition
ECCV 2022 · Table 2
Video segmentation and tracking
595.4SCT-LShifted Chunk Transformer for Spatio-Temporal Representational Learning
NeurIPS 2021 · Table 5
Action recognition
695.3MUFI+fine-tuning
ResNet-101
Boosting Video Representation Learning With Multi-Faceted Integration
CVPR 2021 · Table: Kinetics-400
Action recognition
794.69VATT + GR + CL
VATT-MA-Medium
Scaling Multimodal Pre-Training via Cross-Modality Gradient Harmonization
NeurIPS 2022 · Table 1
Video-language understanding
894.2S2DNet
X3D-S/X3D-XL
Delving into Details: Synopsis-to-Detail Networks for Video Recognition
ECCV 2022 · Table 1
Action recognition
993.1Two-stream KINet
Inception V3
Knowledge Integration Networks for Action Recognition
AAAI 2020 · Comparison with state-of-the-a
Action recognition
1092.04ViT K-centered
ViT-B (ImageNet-1k)
K-Centered Patch Sampling for Efficient Video Recognition
ECCV 2022 · Table 2
Action recognition
1 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.