AI Paper Atlas

Research map / Benchmarks / Something-Something V2

Something-Something V2 leaderboard

93 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on Something-Something V2. Best accuracy: 77.3 (MVD-H, CVPR 2023).

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.

Something-Something V2 · accuracy

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

#accuracy ↑MethodPaperTopic
177.3MVD-H
ViT-Huge
Masked Video Distillation: Rethinking Masked Feature Modeling for Self-Supervised Video Representation Learning
CVPR 2023 · Something-Something V2 compari
Video segmentation and tracking
277VideoMAE V2-g
ViT-g
VideoMAE V2: Scaling Video Masked Autoencoders With Dual Masking
CVPR 2023 · Table 3
Video segmentation and tracking
376.1TubeViT-L
ViT-L
Rethinking Video ViTs: Sparse Video Tubes for Joint Image and Video Learning
CVPR 2023 · Table 4
Video segmentation and tracking
475.5OmniMAE ViT-H
ViT-H
OmniMAE: Single Model Masked Pretraining on Images and Videos
CVPR 2023 · Table: Comparing OmniMAE with
Video segmentation and tracking
575.4VideoMAE ViT-L 32 frames
ViT-L
VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training
NeurIPS 2022 · Table SSV2 SOTA
Video segmentation and tracking
675MaskFeat MViT-L↑312, 40×3 (K600 pre-training)
MViT-L
Masked Feature Prediction for Self-Supervised Visual Pre-Training
CVPR 2022 · Table: Transferring to Somethi
Video segmentation and tracking
774.6ATM ViT-L/14 Merged-2B 16x3x2
ViT-L/14
What Can Simple Arithmetic Operations Do for Temporal Modeling?
ICCV 2023 · Table 3
Action recognition
874.4RLT
ViT-L
Don't Look Twice: Faster Video Transformers with Run-Length Tokenization
NeurIPS 2024 · Training results table
Attention and transformers
974.4MAWS ViT-L
ViT-L, IG-3B
The Effectiveness of MAE Pre-Pretraining for Billion-Scale Pretraining
ICCV 2023 · Table 4
CLIP and zero-shot
1073.8MoMa ViT-L/14 (32 frames)
CLIP ViT-L/14
MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition
ICML 2025 · Table 2
Video segmentation and tracking
71 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

Something-Something V2 · accuracy (val)

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

#accuracy ↑MethodPaperTopic
177.2DejaVid (full learning; w-init-rnd)
VideoMAE V2-g
DejaVid: Encoder-Agnostic Learned Temporal Matching for Video Classification
CVPR 2025 · Table 1
Video segmentation and tracking
273.8MoTED, ViT-L/14, 32 frames
CLIP ViT-L/14
Enhanced Motion-Text Alignment for Image-to-Video Transfer Learning
CVPR 2024 · Table 4
Video-language understanding
372.2ZeroI2V (32x3x1)
CLIP ViT-L/14
ZeroI2V: Zero-Cost Adaptation of Pre-Trained Transformers from Image to Video
ECCV 2024 · Table 5
Video segmentation and tracking
469.5STAN-self-B/16
CLIP-B/16
Revisiting Temporal Modeling for CLIP-Based Image-to-Video Knowledge Transferring
CVPR 2023 · Table: Comparison on Something
Video-language understanding
568.5MLP-3D-L
CycleMLP-B5
MLP-3D: A MLP-Like 3D Architecture With Grouped Time Mixing
CVPR 2022 · Table: Comparisons with the st
Attention and transformers
667.6STAT (64 frames)Shrinking Temporal Attention in Transformers for Video Action Recognition
AAAI 2022 · Table 7
Action recognition
767.4SSA ensemble
ResNet-50
SSAN: Separable Self-Attention Network for Video Representation Learning
CVPR 2021 · Table: Something-Something-V2
Action recognition
867ATA (3×1 views)
TimeSformer
Alignment-guided Temporal Attention for Video Action Recognition
NeurIPS 2022 · Table 2
Action recognition
966.5TEINet-RGB 16f+8f
ResNet-50
TEINet: Towards an Efficient Architecture for Video Recognition
AAAI 2020 · Table: comparison on Something
Action recognition
1065.2MVFNet
ResNet-50, 16 frames, 3 crops, 2 clips
MVFNet: Multi-View Fusion Network for Efficient Video Recognition
AAAI 2021 · Table: Performance and FLOPs c
Action recognition
9 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.