rwightman / pytorch-image-models

PyTorch image models, scripts, pretrained weights

PyTorch Image Models

PyTorch Image Models (TIMM) is a library for state-of-the-art image classification. With this library you can:

Choose from 300+ pre-trained state-of-the-art image classification models.
Train models afresh on research datasets such as ImageNet using provided scripts.
Finetune pre-trained models on your own datasets, including the latest cutting edge models.

Models

MODEL	TRAINED ON	TOP 1 ACCURACY	~FLOPS	YEAR
Vision Transformer	JFT-300M, ImageNet	85.17%	175 Billion	2020
RexNet	ImageNet	81.63%	2 Billion	2020
CSP DarkNet	ImageNet	80.05%	9 Billion	2020
ResNeSt	ImageNet	84.53%	101 Billion	2020
TResNet	ImageNet	83.06%	61 Billion	2020
RegNetX	ImageNet	80.25%	41 Billion	2020
RegNetY	ImageNet	82.01%	4 Billion	2020
EfficientNet Pruned	ImageNet	80.86%	1 Billion	2020
Big Transfer	JFT-300M, ImageNet	84.95%		2019
CSP ResNeXt	ImageNet	80.05%	4 Billion	2019
CSP ResNet	ImageNet	79.57%	6 Billion	2019
AdvProp	ImageNet	85.37%	81 Billion	2019
ESE VovNet	ImageNet	79.31%	9 Billion	2019
Noisy Student	JFT-300M, ImageNet	88.35%	612 Billion	2019
ECAResNet	ImageNet	82.18%	10 Billion	2019
HRNet	ImageNet	79.46%	37 Billion	2019
MixNet	ImageNet	80.47%	1 Billion	2019
TF MixNet	ImageNet	78.78%	689 Million	2019
SelecSLS	ImageNet	78.41%	5 Billion	2019
TF EfficientNet Lite	ImageNet	81.54%	5 Billion	2019
TF EfficientNet	JFT-300M, ImageNet	88.24%	218 Billion	2019
EfficientNet	ImageNet	82.25%	3 Billion	2019
TF MobileNet V3	ImageNet	75.51%	275 Million	2019
MobileNet V3	ImageNet	75.77%	287 Million	2019
SSL ResNet	ImageNet	79.24%	5 Billion	2019
SSL ResNext	ImageNet, YFCC100M	81.84%	47 Billion	2019
SWSL ResNext	ImageNet, IG-1B-Targeted	84.27%	21 Billion	2019
SWSL ResNet	ImageNet	81.14%	5 Billion	2019
TF EfficientNet CondConv	ImageNet	79.33%	370 Million	2019
SPNASNet	ImageNet	74.08%	442 Million	2019
Res2Net	ImageNet	79.19%	11 Billion	2019
Res2NeXt	ImageNet	78.24%	5 Billion	2019
SKResNeXt	ImageNet	80.15%	6 Billion	2019
SKResNet	ImageNet	76.93%	5 Billion	2019
FBNet	ImageNet	75.12%	509 Million	2018
ResNet-D	ImageNet	83.24%	26 Billion	2018
MNASNet	ImageNet	75.45%	415 Million	2018
IG ResNeXt	ImageNet, IG-3.5B-17k	85.42%	197 Billion	2018
Ensemble Adversarial	ImageNet	1.0%	17 Billion	2018
Adversarial Inception v3	ImageNet	77.58%	7 Billion	2018
MobileNet V2	ImageNet	77.28%	889 Million	2018
PNASNet	ImageNet	0.98%	31 Billion	2017
SE ResNet	ImageNet	83.74%	20 Billion	2017
Legacy SENet	ImageNet	81.33%	27 Billion	2017
Legacy SE ResNeXt	ImageNet	80.23%	10 Billion	2017
Legacy SE ResNet	ImageNet	78.67%	15 Billion	2017
Gloun SEResNeXt	ImageNet	80.88%	20 Billion	2017
Gloun SENet	ImageNet	81.23%	27 Billion	2017
SEResNeXt	ImageNet	81.27%	5 Billion	2017
NASNet	ImageNet	82.63%	30 Billion	2017
DLA	ImageNet	79.44%	9 Billion	2017
DPN	ImageNet	80.16%	24 Billion	2017
Gloun ResNeXt	ImageNet	80.63%	20 Billion	2016
ResNeXt	ImageNet	79.79%	5 Billion	2016
Gloun Xception	ImageNet	79.7%	18 Billion	2016
Xception	ImageNet	79.88%	23 Billion	2016
DenseNet	ImageNet	77.36%	10 Billion	2016
Wide ResNet	ImageNet	81.45%	15 Billion	2016
Inception v4	ImageNet	1.01%	16 Billion	2016
Inception ResNet v2	ImageNet	0.95%	17 Billion	2016
Gloun ResNet	ImageNet	81.02%	17 Billion	2015
ResNet	ImageNet	79.29%	7 Billion	2015
Inception v3	ImageNet	77.46%	7 Billion	2015
Gloun Inception v3	ImageNet	78.8%	7 Billion	2015
TF Inception v3	ImageNet	77.87%	7 Billion	2015