Image Classification On Imagenet

评估指标

Hardware Burden

Number of params

Operations per network pass

Top 1 Accuracy

评测结果

各个模型在此基准测试上的表现结果

					Paper Title
CoCa (finetuned)		2100M		91.0%	CoCa: Contrastive Captioners are Image-Text Foundation Models
CoCa (finetuned)	-	2100M	-	91.0%	CoCa: Contrastive Captioners are Image-Text Foundation Models
Model soups (BASIC-L)	-	2440M	-	90.98%	Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time
Model soups (ViT-G/14)	-	1843M	-	90.94%	Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time
DaViT-G	-	1437M	-	90.4%	DaViT: Dual Attention Vision Transformers
DaViT-H	-	362M	-	90.2%	DaViT: Dual Attention Vision Transformers
Meta Pseudo Labels (EfficientNet-L2)	95040G	480M		90.2%	Meta Pseudo Labels
SwinV2-G	-	3000M	-	90.17%	Swin Transformer V2: Scaling Up Capacity and Resolution
MAWS (ViT-6.5B)	-	6500M	-	90.1%	The effectiveness of MAE pre-pretraining for billion-scale pretraining
InternImage-DCNv3-G (M3I Pre-training)	-	3000M	-	90.1%	InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
Florence-CoSwin-H	-	893M	-	90.05%	Florence: A New Foundation Model for Computer Vision
RevCol-H	-	2158M	-	90.0%	Reversible Column Networks
Meta Pseudo Labels (EfficientNet-B6-Wide)	-	390M	-	90%	Meta Pseudo Labels
MAWS (ViT-2B)	-	2000M	-	89.8%	The effectiveness of MAE pre-pretraining for billion-scale pretraining
EVA	-	1000M	-	89.7%	EVA: Exploring the Limits of Masked Visual Representation Learning at Scale
M3I Pre-training (InternImage-H)	-	-	-	89.6%	Towards All-in-one Pre-training via Maximizing Multi-modal Mutual Information
ViT-L/16 (384res, distilled from ViT-22B)	-	307M	-	89.6%	Scaling Vision Transformers to 22 Billion Parameters
InternImage-H	-	1080M	-	89.6%	InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
MaxViT-XL (512res, JFT)	-	-	-	89.53%	MaxViT: Multi-Axis Vision Transformer
AIMv2-3B (448 res)	-	-	-	89.5%	Multimodal Autoregressive Pre-training of Large Vision Encoders

0 of 1058 row(s) selected.