kerascv

Large-scale image classification models on Keras

This is a collection of large-scale image classification models. Many of them are pretrained on ImageNet-1K dataset and loaded automatically during use. All pretrained models require the same ordinary normalization. Scripts for training/evaluating/converting models are in the imgclsmob repo.

List of implemented models

• AlexNet ('One weird trick for parallelizing convolutional neural networks')
• VGG/BN-VGG ('Very Deep Convolutional Networks for Large-Scale Image Recognition')
• ResNet ('Deep Residual Learning for Image Recognition')
• PreResNet ('Identity Mappings in Deep Residual Networks')
• ResNeXt ('Aggregated Residual Transformations for Deep Neural Networks')
• SENet/SE-ResNet/SE-PreResNet/SE-ResNeXt ('Squeeze-and-Excitation Networks')
• DenseNet ('Densely Connected Convolutional Networks')
• DarkNet Ref/Tiny/19 ('Darknet: Open source neural networks in c')
• DarkNet-53 ('YOLOv3: An Incremental Improvement')
• SqueezeNet/SqueezeResNet ('SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size')
• SqueezeNext ('SqueezeNext: Hardware-Aware Neural Network Design')
• ShuffleNet ('ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices')
• ShuffleNetV2 ('ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design')
• MENet ('Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications')
• MobileNet ('MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications')
• FD-MobileNet ('FD-MobileNet: Improved MobileNet with A Fast Downsampling Strategy')
• MobileNetV2 ('MobileNetV2: Inverted Residuals and Linear Bottlenecks')
• IGCV3 ('IGCV3: Interleaved Low-Rank Group Convolutions for Efficient Deep Neural Networks')
• MnasNet ('MnasNet: Platform-Aware Neural Architecture Search for Mobile')
• EfficientNet ('EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks')

Installation

To use the models in your project, simply install the kerascv package with desired backend. For example for MXNet backend:

pip install mxnet>=1.2.1 keras-mxnet kerascv

Or if you prefer TensorFlow backend:

pip install tensorflow kerascv

To enable/disable different hardware supports, check out installation instruction for the corresponding backend.

After installation check that the backend field is set to the correct value in the file ~/.keras/keras.json. It is also preferable to set the value of the image_data_format field to channels_first in the case of using the MXNet backend.

Usage

Example of using a pretrained ResNet-18 model (for channels_first data format):

from kerascv.model_provider import get_model as kecv_get_model
import numpy as np

net = kecv_get_model("resnet18", pretrained=True)
x = np.zeros((1, 3, 224, 224), np.float32)
y = net.predict(x)

Pretrained models (ImageNet-1K)

Some remarks:

• All quality values are estimated with MXNet backend.
• Top1/Top5 are the standard 1-crop Top-1/Top-5 errors (in percents) on the validation subset of the ImageNet-1K dataset.
• FLOPs/2 is the number of FLOPs divided by two to be similar to the number of MACs.
• Remark Converted from GL model means that the model was trained on MXNet/Gluon and then converted to Keras.

Model	Top1	Top5	Params	FLOPs/2	Remarks
AlexNet	40.47	17.88	62,378,344	1,132.33M	Converted from GL model (log)
AlexNet-b	41.08	18.53	61,100,840	714.83M	Converted from GL model (log)
ZFNet	39.56	17.15	62,357,608	1,170.33M	Converted from GL model (log)
ZFNet-b	36.30	14.83	107,627,624	2,479.13M	Converted from GL model (log)
VGG-11	29.59	10.16	132,863,336	7,615.87M	Converted from GL model (log)
VGG-13	28.37	9.50	133,047,848	11,317.65M	Converted from GL model (log)
VGG-16	26.61	8.32	138,357,544	15,480.10M	Converted from GL model (log)
VGG-19	25.58	7.67	143,667,240	19,642.55M	Converted from GL model (log)
BN-VGG-11	28.55	9.34	132,866,088	7,630.21M	Converted from GL model (log)
BN-VGG-13	27.68	8.87	133,050,792	11,341.62M	Converted from GL model (log)
BN-VGG-16	25.50	7.57	138,361,768	15,506.38M	Converted from GL model (log)
BN-VGG-19	23.91	6.89	143,672,744	19,671.15M	Converted from GL model (log)
BN-VGG-11b	29.24	9.75	132,868,840	7,630.72M	Converted from GL model (log)
BN-VGG-13b	29.48	10.16	133,053,736	11,342.14M	From dmlc/gluon-cv (log)
BN-VGG-16b	26.88	8.65	138,365,992	15,507.20M	From dmlc/gluon-cv (log)
BN-VGG-19b	25.65	8.14	143,678,248	19,672.26M	From dmlc/gluon-cv (log)
ResNet-10	34.59	13.85	5,418,792	894.04M	Converted from GL model (log)
ResNet-12	33.43	13.03	5,492,776	1,126.25M	Converted from GL model (log)
ResNet-14	32.18	12.20	5,788,200	1,357.94M	Converted from GL model (log)
ResNet-BC-14b	30.25	11.16	10,064,936	1,479.12M	Converted from GL model (log)
ResNet-16	30.23	10.88	6,968,872	1,589.34M	Converted from GL model (log)
ResNet-18 x0.25	39.30	17.41	3,937,400	270.94M	Converted from GL model (log)
ResNet-18 x0.5	33.40	12.83	5,804,296	608.70M	Converted from GL model (log)
ResNet-18 x0.75	29.98	10.66	8,476,056	1,129.45M	Converted from GL model (log)
ResNet-18	28.08	9.52	11,689,512	1,820.41M	Converted from GL model (log)
ResNet-26	26.12	8.37	17,960,232	2,746.79M	Converted from GL model (log)
ResNet-BC-26b	24.85	7.59	15,995,176	2,356.67M	Converted from GL model (log)
ResNet-34	24.53	7.44	21,797,672	3,672.68M	Converted from GL model (log)
ResNet-BC-38b	23.48	6.72	21,925,416	3,234.21M	Converted from GL model (log)
ResNet-50	22.14	6.04	25,557,032	3,877.95M	Converted from GL model (log)
ResNet-50b	22.06	6.10	25,557,032	4,110.48M	Converted from GL model (log)
ResNet-101	21.64	5.99	44,549,160	7,597.95M	From dmlc/gluon-cv (log)
ResNet-101b	20.25	5.11	44,549,160	7,830.48M	Converted from GL model (log)
ResNet-152	20.74	5.35	60,192,808	11,321.85M	From dmlc/gluon-cv (log)
ResNet-152b	19.63	4.79	60,192,808	11,554.38M	Converted from GL model (log)
PreResNet-10	34.65	14.01	5,417,128	894.19M	Converted from GL model (log)
PreResNet-12	33.56	13.22	5,491,112	1,126.40M	Converted from GL model (log)
PreResNet-14	32.29	12.19	5,786,536	1,358.09M	Converted from GL model (log)
PreResNet-BC-14b	30.66	11.51	10,057,384	1,476.62M	Converted from GL model (log)
PreResNet-16	30.21	10.81	6,967,208	1,589.49M	Converted from GL model (log)
PreResNet-18 x0.25	39.63	17.78	3,935,960	270.93M	Converted from GL model (log)
PreResNet-18 x0.5	33.67	13.19	5,802,440	608.73M	Converted from GL model (log)
PreResNet-18 x0.75	29.95	10.68	8,473,784	1,129.51M	Converted from GL model (log)
PreResNet-18	28.16	9.52	11,687,848	1,820.56M	Converted from GL model (log)
PreResNet-26	26.02	8.34	17,958,568	2,746.94M	Converted from GL model (log)
PreResNet-BC-26b	25.20	7.86	15,987,624	2,354.16M	Converted from GL model (log)
PreResNet-34	24.55	7.51	21,796,008	3,672.83M	Converted from GL model (log)
PreResNet-BC-38b	22.65	6.33	21,917,864	3,231.70M	Converted from GL model (log)
PreResNet-50	22.26	6.20	25,549,480	3,875.44M	Converted from GL model (log)
PreResNet-50b	22.35	6.32	25,549,480	4,107.97M	Converted from GL model (log)
PreResNet-101	21.43	5.75	44,541,608	7,595.44M	From dmlc/gluon-cv (log)
PreResNet-101b	20.84	5.40	44,541,608	7,827.97M	Converted from GL model (log)
PreResNet-152	20.69	5.31	60,185,256	11,319.34M	From dmlc/gluon-cv (log)
PreResNet-152b	19.89	5.00	60,185,256	11,551.87M	Converted from GL model (log)
PreResNet-200b	21.09	5.64	64,666,280	15,068.63M	From tornadomeet/ResNet (log)
PreResNet-269b	20.71	5.56	102,065,832	20,101.11M	From soeaver/mxnet-model (log)
ResNeXt-14 (16x4d)	31.65	12.24	7,127,336	1,045.77M	Converted from GL model (log)
ResNeXt-14 (32x2d)	32.15	12.46	7,029,416	1,031.32M	Converted from GL model (log)
ResNeXt-14 (32x4d)	29.95	11.10	9,411,880	1,603.46M	Converted from GL model (log)
ResNeXt-26 (32x2d)	26.34	8.50	9,924,136	1,461.06M	Converted from GL model (log)
ResNeXt-26 (32x4d)	23.91	7.20	15,389,480	2,488.07M	Converted from GL model (log)
ResNeXt-50 (32x4d)	20.64	5.46	25,028,904	4,255.86M	From dmlc/gluon-cv (log)
ResNeXt-101 (32x4d)	19.62	4.92	44,177,704	8,003.45M	From dmlc/gluon-cv (log)
ResNeXt-101 (64x4d)	19.28	4.83	83,455,272	15,500.27M	From dmlc/gluon-cv (log)
SE-ResNet-10	33.55	13.29	5,463,332	894.27M	Converted from GL model (log)
SE-ResNet-18	27.95	9.20	11,778,592	1,820.88M	Converted from GL model (log)
SE-ResNet-26	25.42	8.03	18,093,852	2,747.49M	Converted from GL model (log)
SE-ResNet-BC-26b	23.44	6.82	17,395,976	2,359.58M	Converted from GL model (log)
SE-ResNet-BC-38b	21.44	5.75	24,026,616	3,238.58M	Converted from GL model (log)
SE-ResNet-50	22.50	6.43	28,088,024	3,880.49M	From Cadene/pretrained...pytorch (log)
SE-ResNet-50b	20.58	5.33	28,088,024	4,115.78M	Converted from GL model (log)
SE-ResNet-101	21.92	5.88	49,326,872	7,602.76M	From Cadene/pretrained...pytorch (log)
SE-ResNet-152	21.46	5.77	66,821,848	11,328.52M	From Cadene/pretrained...pytorch (log)
SE-PreResNet-10	33.60	13.06	5,461,668	894.42M	Converted from GL model (log)
SE-PreResNet-18	27.67	9.38	11,776,928	1,821.03M	Converted from GL model (log)
SE-PreResNet-BC-26b	22.95	6.36	17,388,424	2,357.07M	Converted from GL model (log)
SE-PreResNet-BC-38b	21.42	5.63	24,019,064	3,236.07M	Converted from GL model (log)
SE-ResNeXt-50 (32x4d)	20.03	5.05	27,559,896	4,261.16M	From dmlc/gluon-cv (log)
SE-ResNeXt-101 (32x4d)	19.07	4.60	48,955,416	8,012.73M	From dmlc/gluon-cv (log)
SE-ResNeXt-101 (64x4d)	18.98	4.66	88,232,984	15,509.54M	From dmlc/gluon-cv (log)
SENet-16	25.34	8.06	31,366,168	5,081.30M	Converted from GL model (log)
SENet-28	21.68	5.91	36,453,768	5,732.71M	Converted from GL model (log)
SENet-154	18.83	4.65	115,088,984	20,745.78M	From Cadene/pretrained...pytorch (log)
DenseNet-121	23.23	6.84	7,978,856	2,872.13M	Converted from GL model (log)
DenseNet-161	22.39	6.18	28,681,000	7,793.16M	From dmlc/gluon-cv (log)
DenseNet-169	22.09	6.05	14,149,480	3,403.89M	Converted from GL model (log)
DenseNet-201	22.69	6.35	20,013,928	4,347.15M	From dmlc/gluon-cv (log)
DarkNet Tiny	40.31	17.46	1,042,104	500.85M	Converted from GL model (log)
DarkNet Ref	37.99	16.68	7,319,416	367.59M	Converted from GL model (log)
DarkNet-53	21.43	5.56	41,609,928	7,133.86M	From dmlc/gluon-cv (log)
SqueezeNet v1.0	39.17	17.56	1,248,424	823.67M	Converted from GL model (log)
SqueezeNet v1.1	39.08	17.39	1,235,496	352.02M	Converted from GL model (log)
SqueezeResNet v1.0	39.40	17.80	1,248,424	823.67M	Converted from GL model (log)
SqueezeResNet v1.1	39.82	17.84	1,235,496	352.02M	Converted from GL model (log)
1.0-SqNxt-23	42.28	18.62	724,056	287.28M	Converted from GL model (log)
1.0-SqNxt-23v5	40.38	17.57	921,816	285.82M	Converted from GL model (log)
1.5-SqNxt-23	34.59	13.30	1,511,824	552.39M	Converted from GL model (log)
1.5-SqNxt-23v5	33.56	12.84	1,953,616	550.97M	Converted from GL model (log)
2.0-SqNxt-23	30.15	10.66	2,583,752	898.48M	Converted from GL model (log)
2.0-SqNxt-23v5	29.40	10.28	3,366,344	897.60M	Converted from GL model (log)
ShuffleNet x0.25 (g=1)	62.00	36.76	209,746	12.35M	Converted from GL model (log)
ShuffleNet x0.25 (g=3)	61.32	36.15	305,902	13.09M	Converted from GL model (log)
ShuffleNet x0.5 (g=1)	46.21	22.38	534,484	41.16M	Converted from GL model (log)
ShuffleNet x0.5 (g=3)	43.82	20.60	718,324	41.70M	Converted from GL model (log)
ShuffleNet x0.75 (g=1)	39.24	16.75	975,214	86.42M	Converted from GL model (log)
ShuffleNet x0.75 (g=3)	37.81	16.09	1,238,266	85.82M	Converted from GL model (log)
ShuffleNet x1.0 (g=1)	34.41	13.50	1,531,936	148.13M	Converted from GL model (log)
ShuffleNet x1.0 (g=2)	33.97	13.32	1,733,848	147.60M	Converted from GL model (log)
ShuffleNet x1.0 (g=3)	33.96	13.29	1,865,728	145.46M	Converted from GL model (log)
ShuffleNet x1.0 (g=4)	33.83	13.10	1,968,344	143.33M	Converted from GL model (log)
ShuffleNet x1.0 (g=8)	33.64	13.20	2,434,768	150.76M	Converted from GL model (log)
ShuffleNetV2 x0.5	40.76	18.40	1,366,792	43.31M	Converted from GL model (log)
ShuffleNetV2 x1.0	31.02	11.33	2,278,604	149.72M	Converted from GL model (log)
ShuffleNetV2 x1.5	27.32	9.27	4,406,098	320.77M	Converted from GL model (log)
ShuffleNetV2 x2.0	25.77	8.22	7,601,686	595.84M	Converted from GL model (log)
ShuffleNetV2b x0.5	39.81	17.83	1,366,792	43.31M	Converted from GL model (log)
ShuffleNetV2b x1.0	30.38	11.01	2,279,760	150.62M	Converted from GL model (log)
ShuffleNetV2b x1.5	26.89	8.80	4,410,194	323.98M	Converted from GL model (log)
ShuffleNetV2b x2.0	25.18	8.10	7,611,290	603.37M	Converted from GL model (log)
108-MENet-8x1 (g=3)	43.61	20.31	654,516	42.68M	Converted from GL model (log)
128-MENet-8x1 (g=4)	42.08	19.14	750,796	45.98M	Converted from GL model (log)
160-MENet-8x1 (g=8)	43.47	20.28	850,120	45.63M	Converted from GL model (log)
228-MENet-12x1 (g=3)	33.85	12.88	1,806,568	152.93M	Converted from GL model (log)
256-MENet-12x1 (g=4)	32.22	12.17	1,888,240	150.65M	Converted from GL model (log)
348-MENet-12x1 (g=3)	27.85	9.36	3,368,128	312.00M	Converted from GL model (log)
352-MENet-12x1 (g=8)	31.29	11.67	2,272,872	157.35M	Converted from GL model (log)
456-MENet-24x1 (g=3)	25.00	7.80	5,304,784	567.90M	Converted from GL model (log)
MobileNet x0.25	45.80	22.17	470,072	44.09M	Converted from GL model (log)
MobileNet x0.5	33.94	13.30	1,331,592	155.42M	Converted from GL model (log)
MobileNet x0.75	29.85	10.51	2,585,560	333.99M	Converted from GL model (log)
MobileNet x1.0	26.43	8.66	4,231,976	579.80M	Converted from GL model (log)
FD-MobileNet x0.25	55.42	30.52	383,160	12.95M	Converted from GL model (log)
FD-MobileNet x0.5	42.61	19.69	993,928	41.84M	Converted from GL model (log)
FD-MobileNet x0.75	37.90	16.01	1,833,304	86.68M	Converted from GL model (log)
FD-MobileNet x1.0	33.80	13.12	2,901,288	147.46M	Converted from GL model (log)
MobileNetV2 x0.25	48.06	24.12	1,516,392	34.24M	Converted from GL model (log)
MobileNetV2 x0.5	35.63	14.43	1,964,736	100.13M	Converted from GL model (log)
MobileNetV2 x0.75	29.76	10.44	2,627,592	198.50M	Converted from GL model (log)
MobileNetV2 x1.0	26.76	8.64	3,504,960	329.36M	Converted from GL model (log)
MobileNetV3 L/224/1.0	24.63	7.69	5,481,752	227.09M	From dmlc/gluon-cv (log)
IGCV3 x0.25	53.41	28.29	1,534,020	41.29M	Converted from GL model (log)
IGCV3 x0.5	39.39	17.04	1,985,528	111.12M	Converted from GL model (log)
IGCV3 x0.75	30.71	10.97	2,638,084	210.95M	Converted from GL model (log)
IGCV3 x1.0	27.72	8.99	3,491,688	340.79M	Converted from GL model (log)
MnasNet-B1	25.76	8.00	4,383,312	326.30M	From rwightman/pyt...models (log)
MnasNet-A1	25.02	7.55	3,887,038	326.07M	From rwightman/pyt...models (log)
EfficientNet-B0	24.50	7.22	5,288,548	414.31M	Converted from GL model (log)
EfficientNet-B1	22.89	6.26	7,794,184	732.54M	Converted from GL model (log)
EfficientNet-B0b	22.95	6.69	5,288,548	414.31M	From rwightman/pyt...models (log)
EfficientNet-B1b	20.97	5.64	7,794,184	732.54M	From rwightman/pyt...models (log)
EfficientNet-B2b	19.93	5.16	9,109,994	1,051.98M	From rwightman/pyt...models (log)
EfficientNet-B3b	18.59	4.31	12,233,232	1,928.55M	From rwightman/pyt...models (log)
EfficientNet-B4b	17.24	3.76	19,341,616	4,607.46M	From rwightman/pyt...models (log)
EfficientNet-B5b	16.39	3.34	30,389,784	10,695.20M	From rwightman/pyt...models (log)
EfficientNet-B6b	15.96	3.12	43,040,704	19,796.24M	From rwightman/pyt...models (log)
EfficientNet-B7b	15.70	3.11	66,347,960	39,010.98M	From rwightman/pyt...models (log)