Action Classification On Kinetics 700

评估指标

Top-1 Accuracy

Top-5 Accuracy

评测结果

各个模型在此基准测试上的表现结果

			Paper Title	Repository
InternVideo2-6B	85.9	-	InternVideo2: Scaling Foundation Models for Multimodal Video Understanding
InternVideo2-1B	85.4	-	InternVideo2: Scaling Foundation Models for Multimodal Video Understanding
InternVideo-T	84.0	-	InternVideo: General Video Foundation Models via Generative and Discriminative Learning
TubeViT-L	83.8	96.6	Rethinking Video ViTs: Sparse Video Tubes for Joint Image and Video Learning
UMT-L (ViT-L/16)	83.6	96.7	Unmasked Teacher: Towards Training-Efficient Video Foundation Models
MTV-H (WTS 60M)	83.4	96.2	Multiview Transformers for Video Recognition
EVA	82.9%	-	EVA: Exploring the Limits of Masked Visual Representation Learning at Scale
UniFormerV2-L	82.7	96.2	UniFormerV2: Spatiotemporal Learning by Arming Image ViTs with Video UniFormer	-
CoCa (finetuned)	82.7	-	CoCa: Contrastive Captioners are Image-Text Foundation Models
CoCa (frozen)	81.1	-	CoCa: Contrastive Captioners are Image-Text Foundation Models
Hiera-H (no extra data)	81.1	-	Hiera: A Hierarchical Vision Transformer without the Bells-and-Whistles
MaskFeat (no extra data, MViT-L)	80.4	95.7	Masked Feature Prediction for Self-Supervised Visual Pre-Training
mPLUG-2	80.4	94.9	mPLUG-2: A Modularized Multi-modal Foundation Model Across Text, Image and Video
AIM (CLIP ViT-L/14, 32x224)	80.4	-	AIM: Adapting Image Models for Efficient Video Action Recognition
CoVeR (JFT-3B)	79.8	94.9	Co-training Transformer with Videos and Images Improves Action Recognition	-
MViTv2-L (ImageNet-21k pretrain)	79.4	94.9	MViTv2: Improved Multiscale Vision Transformers for Classification and Detection
MoViNet-A6	79.4	-	MViTv2: Improved Multiscale Vision Transformers for Classification and Detection
CoVeR (JFT-300M)	78.5	94.2	Co-training Transformer with Videos and Images Improves Action Recognition	-
MViTv2-B	76.6	93.2	MViTv2: Improved Multiscale Vision Transformers for Classification and Detection
MoViNet-A6	72.3	-	MoViNets: Mobile Video Networks for Efficient Video Recognition

0 of 36 row(s) selected.