imagenet 测试精度合集

目录

ConvNeXt：新一代卷积网络，还不是ViT的时代（代码开源）

GENet

regnet

poolformer

CoTNet

ParNet

Convolutional-MLPs

renet精度

repvgg

ConvMixer

hs-resnet50

RedNet

nfnet

SimAM

---

ConvNeXt：新一代卷积网络，还不是ViT的时代（代码开源）

ConvNeXt：新一代卷积网络，还不是ViT的时代（代码开源）

GENet

GPU端吊打RegNet、EfficientNet的强悍担当：GENet_喜欢打酱油的老鸟的博客-CSDN博客_efficientnet网络

regnet

regnet_AI视觉网奇的博客-CSDN博客

poolformer

https://github.com/sail-sg/poolformer

模型46M,精度77.2

模型82M,精度80.3

Model	#params	Image resolution	Top1 Acc	Download
poolformer_s12	12M	224	77.2	here
poolformer_s24	21M	224	80.3	here
poolformer_s36	31M	224	81.4	here
poolformer_m36	56M	224	82.1	here
poolformer_m48	73M	224	82.5	here

CoTNet

京东开源的：

https://github.com/JDAI-CV/CoTNet

CoTNet-50	224	22.2M	3.3	81.3	95.6	GoogleDrive / Baidu
CoTNeXt-50	224	30.1M	4.3	82.1	95.9	GoogleDrive / Baidu
SE-CoTNetD-50	224	23.1M	4.1	81.6	95.8	GoogleDrive / Baidu
CoTNet-101	224	38.3M	6.1	82.8	96.2	GoogleDrive / Baidu
CoTNeXt-101	224	53.4M	8.2	83.2	96.4	GoogleDrive / Baidu
SE-CoTNetD-101	224	40.9M	8.5	83.2	96.5	GoogleDrive / Baidu
SE-CoTNetD-152	224	55.8M	17.0	84.0	97.0	GoogleDrive / Baidu
SE-CoTNetD-152	320	55.8M	26.5	84.6	97.1	GoogleDrive / Baidu

ParNet

12层也能媲美ResNet

代码好像已开源，没有预训练：

https://github.com/imankgoyal/NonDeepNetworks

本机整理的代码：

F:\project\cls\NonDeepNetworks

simpnet.py

demo.py

trt_demo.py

onnx2trt.py

Convolutional-MLPs

https://github.com/SHI-Labs/Convolutional-MLPs

36m模型，精度76.8，比resent-50稍微好一点。有下载链接:

Dataset	Model	Top-1 Accuracy	# Params	MACs
ImageNet	ConvMLP-S	76.8%	9.0M	2.4G
	ConvMLP-M	79.0%	17.4M	3.9G
	ConvMLP-L	80.2%	42.7M	9.9G

renet精度

Network	Top-1 error	Top-5 error
ResNet-18	69.57	89.24
ResNet-34	73.27	91.26
ResNet-50	75.99	92.98
ResNet-101	77.56	93.79
ResNet-152	77.84	93.84
ResNet-200	78.34	94.21

repvgg

有预训练地址：

GitHub - DingXiaoH/RepVGG: RepVGG: Making VGG-style ConvNets Great Again

提供了百度网盘地址：

网络结构：

https://github.com/megvii-research/basecls/tree/main/zoo/public/repvgg

ConvMixer

讲解：

ConvMixer来了！单挑ResNet、ViT、MLP-Mixer的简单模型

代码：

GitHub - tmp-iclr/convmixer

hs-resnet50

RedNet

https://github.com/d-li14/involution

Model	Params(M)	FLOPs(G)	Top-1 (%)	Top-5 (%)	Config	Download
RedNet-26	9.23(32.8%↓)	1.73(29.2%↓)	75.96	93.19	config	model | log
RedNet-38	12.39(36.7%↓)	2.22(31.3%↓)	77.48	93.57	config	model | log
RedNet-50	15.54(39.5%↓)	2.71(34.1%↓)	78.35	94.13	config	model | log
RedNet-101	25.65(42.6%↓)	4.74(40.5%↓)	78.92	94.35	config	model | log
RedNet-152	33.99(43.5%↓)	6.79(41.4%↓)	79.12	94.38	config	model | log

nfnet

我发现一个问题，第一层只有16个通道

deepmind-research/nfnets at master · deepmind/deepmind-research · GitHub

• 论文链接：https://arxiv.org/abs/2102.06171

• DeepMind 还放出了模型的实现：https://github.com/deepmind/deepmind-research/tree/master/nfnets

SimAM

GitHub - ZjjConan/SimAM: The official pytorch implemention of our ICML paper "SimAM: A Simple, Parameter-Free Attention Module for Convolutional Neural Networks".

Model	Parameters	FLOPs	Top-1(%)	Top-5(%)
SimAM-R18	11.69 M	1.82 G	71.31	89.88
SimAM-R32	21.80 M	3.67 G	74.49	92.02
SimAM-R50	25.56 M	4.11 G	77.45	93.66
SimAM-R101	44.55 M	7.83 G	78.65	94.11
SimAM-RX50 (32x4d)	25.03 M	4.26 G	78.00	93.93
SimAM-MV2	3.50 M	0.31 G	72.36	90.74