Image Classification On Imagenet

المقاييس

Hardware Burden

Number of params

Operations per network pass

Top 1 Accuracy

النتائج

نتائج أداء النماذج المختلفة على هذا المعيار القياسي

اسم النموذج	Hardware Burden	Number of params	Operations per network pass	Top 1 Accuracy	Paper Title	Repository
Xception	87G	22.855952M	0.838G	79%	Xception: Deep Learning with Depthwise Separable Convolutions
ResNet-101	-	40M	-	78.25%	Deep Residual Learning for Image Recognition
CvT-13 (384 res)	-	20M	-	83%	CvT: Introducing Convolutions to Vision Transformers
DenseNet-201	-	-	-	77.42%	Densely Connected Convolutional Networks
ConViT-Ti+	-	10M	-	76.7%	ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases
ViT-B/16-SAM	-	87M	-	79.9%	When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations
PVTv2-B2	-	25.4M	-	82%	PVT v2: Improved Baselines with Pyramid Vision Transformer
ConvFormer-S36 (224 res, 21K)	-	40M	-	85.4%	MetaFormer Baselines for Vision
MambaVision-L2	-	241.5M	-	85.3%	MambaVision: A Hybrid Mamba-Transformer Vision Backbone
FBNetV5-AC-CLS	-	-	-	78.4%	FBNetV5: Neural Architecture Search for Multiple Tasks in One Run	-
HCGNet-B	-	12.9M	-	78.5%	Gated Convolutional Networks with Hybrid Connectivity for Image Classification
ConViT-S+	-	48M	-	82.2%	ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases
ResMLP-B24 + STD	-	122.6M	-	82.4%	Spatial-Channel Token Distillation for Vision MLPs
DeiT-S with iRPE-K	-	22M	-	80.9%	Rethinking and Improving Relative Position Encoding for Vision Transformer
BoTNet T6	-	53.9M	-	84%	Bottleneck Transformers for Visual Recognition
ViT-B @224 (DeiT-III + AugSub)	-	86.6M	-	84.2%	Masking Augmentation for Supervised Learning
MaxViT-B (224res)	-	120M	-	84.94%	MaxViT: Multi-Axis Vision Transformer
BiFormer-S* (IN1k ptretrain)	-	-	-	84.3%	BiFormer: Vision Transformer with Bi-Level Routing Attention
CeiT-T	-	6.4M	-	76.4%	Incorporating Convolution Designs into Visual Transformers
ResNet-101 (AutoMix)	-	44.6M	-	80.98%	AutoMix: Unveiling the Power of Mixup for Stronger Classifiers

0 of 1058 row(s) selected.