gluoncv2 0.0.32

Image classification models for Gluon

Image classification models on MXNet/Gluon

This is a collection of image classification models. Many of them are pretrained on ImageNet-1K and CIFAR-10 datasets 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')
• ZFNet ('Visualizing and Understanding Convolutional Networks')
• VGG/BN-VGG ('Very Deep Convolutional Networks for Large-Scale Image Recognition')
• BN-Inception ('Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift')
• 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')
• IBN-ResNet/IBN-ResNeXt/IBN-DenseNet ('Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net')
• AirNet/AirNeXt ('Attention Inspiring Receptive-Fields Network for Learning Invariant Representations')
• BAM-ResNet ('BAM: Bottleneck Attention Module')
• CBAM-ResNet ('CBAM: Convolutional Block Attention Module')
• ResAttNet ('Residual Attention Network for Image Classification')
• PyramidNet ('Deep Pyramidal Residual Networks')
• DiracNetV2 ('DiracNets: Training Very Deep Neural Networks Without Skip-Connections')
• DenseNet ('Densely Connected Convolutional Networks')
• CondenseNet ('CondenseNet: An Efficient DenseNet using Learned Group Convolutions')
• SparseNet ('Sparsely Aggregated Convolutional Networks')
• PeleeNet ('Pelee: A Real-Time Object Detection System on Mobile Devices')
• WRN ('Wide Residual Networks')
• DRN-C/DRN-D ('Dilated Residual Networks')
• DPN ('Dual Path Networks')
• DarkNet Ref/Tiny/19 ('Darknet: Open source neural networks in c')
• DarkNet-53 ('YOLOv3: An Incremental Improvement')
• ChannelNet ('ChannelNets: Compact and Efficient Convolutional Neural Networks via Channel-Wise Convolutions')
• MSDNet ('Multi-Scale Dense Networks for Resource Efficient Image Classification')
• FishNet ('FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction')
• 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')
• DARTS ('DARTS: Differentiable Architecture Search')
• Xception ('Xception: Deep Learning with Depthwise Separable Convolutions')
• InceptionV3 ('Rethinking the Inception Architecture for Computer Vision')
• InceptionV4/InceptionResNetV2 ('Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning')
• PolyNet ('PolyNet: A Pursuit of Structural Diversity in Very Deep Networks')
• NASNet ('Learning Transferable Architectures for Scalable Image Recognition')
• PNASNet ('Progressive Neural Architecture Search')

Installation

To use the models in your project, simply install the gluoncv2 package with mxnet:

pip install gluoncv2 mxnet>=1.2.1

To enable different hardware supports such as GPUs, check out MXNet variants. For example, you can install with CUDA-9.2 supported MXNet:

pip install gluoncv2 mxnet-cu92>=1.2.1

Usage

Example of using the pretrained ResNet-18 model:

from gluoncv2.model_provider import get_model as glcv2_get_model
import mxnet as mx

net = glcv2_get_model("resnet18", pretrained=True)
x = mx.nd.zeros((1, 3, 224, 224), ctx=mx.cpu())
y = net(x)

Pretrained models

Imagenet-1K

Some remarks:

• 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.
• ResNet/PreResNet with b-suffix is a version of the networks with the stride in the second convolution of the bottleneck block. Respectively a network without b-suffix has the stride in the first convolution.
• ResNet/PreResNet models do not use biases in convolutions at all.
• CondenseNet models are only so-called converted versions.
• ShuffleNetV2/ShuffleNetV2b/ShuffleNetV2c are different implementations of the same architecture.

Model	Top1	Top5	Params	FLOPs/2	Remarks
AlexNet	44.12	21.26	61,100,840	714.83M	From dmlc/gluon-cv (log)
VGG-11	31.91	11.76	132,863,336	7,615.87M	From dmlc/gluon-cv (log)
VGG-13	31.06	11.12	133,047,848	11,317.65M	From dmlc/gluon-cv (log)
VGG-16	26.78	8.69	138,357,544	15,480.10M	From dmlc/gluon-cv (log)
VGG-19	25.88	8.23	143,667,240	19,642.55M	From dmlc/gluon-cv (log)
BN-VGG-11b	30.34	10.57	132,868,840	7,630.72M	From dmlc/gluon-cv (log)
BN-VGG-13b	29.48	10.16	133,053,736	11,342.14M	From dmlc/gluon-cv (log)
BN-VGG-16b	26.89	8.65	138,365,992	15,507.20M	From dmlc/gluon-cv (log)
BN-VGG-19b	25.66	8.15	143,678,248	19,672.26M	From dmlc/gluon-cv (log)
BN-Inception	25.09	7.76	11,295,240	2,048.06M	From Cadene/pretrained...pytorch (log)
ResNet-10	37.09	15.55	5,418,792	894.04M	Training (log)
ResNet-12	35.86	14.46	5,492,776	1,126.25M	Training (log)
ResNet-14	32.85	12.41	5,788,200	1,357.94M	Training (log)
ResNet-16	30.68	11.10	6,968,872	1,589.34M	Training (log)
ResNet-18 x0.25	49.16	24.45	831,096	137.32M	Training (log)
ResNet-18 x0.5	36.54	14.96	3,055,880	486.49M	Training (log)
ResNet-18 x0.75	33.25	12.54	6,675,352	1,047.53M	Training (log)
ResNet-18	28.09	9.51	11,689,512	1,820.41M	Training (log)
ResNet-34	25.34	7.92	21,797,672	3,672.68M	From dmlc/gluon-cv (log)
ResNet-50	22.65	6.41	25,557,032	3,877.95M	From dmlc/gluon-cv (log)
ResNet-50b	22.32	6.18	25,557,032	4,110.48M	From dmlc/gluon-cv (log)
ResNet-101	21.66	5.99	44,549,160	7,597.95M	From dmlc/gluon-cv (log)
ResNet-101b	20.79	5.39	44,549,160	7,830.48M	From dmlc/gluon-cv (log)
ResNet-152	20.76	5.35	60,192,808	11,321.85M	From dmlc/gluon-cv (log)
ResNet-152b	20.31	5.25	60,192,808	11,554.38M	From dmlc/gluon-cv (log)
PreResNet-18	28.16	9.51	11,687,848	1,820.56M	Training (log)
PreResNet-34	25.88	8.11	21,796,008	3,672.83M	From dmlc/gluon-cv (log)
PreResNet-50	23.39	6.68	25,549,480	3,875.44M	From dmlc/gluon-cv (log)
PreResNet-50b	23.16	6.64	25,549,480	4,107.97M	From dmlc/gluon-cv (log)
PreResNet-101	21.45	5.75	44,541,608	7,595.44M	From dmlc/gluon-cv (log)
PreResNet-101b	21.73	5.88	44,541,608	7,827.97M	From dmlc/gluon-cv (log)
PreResNet-152	20.70	5.32	60,185,256	11,319.34M	From dmlc/gluon-cv (log)
PreResNet-152b	21.00	5.75	60,185,256	11,551.87M	From dmlc/gluon-cv (log)
PreResNet-200b	21.10	5.64	64,666,280	15,068.63M	From tornadomeet/ResNet (log)
ResNeXt-101 (32x4d)	21.32	5.79	44,177,704	8,003.45M	From Cadene/pretrained...pytorch (log)
ResNeXt-101 (64x4d)	20.60	5.41	83,455,272	15,500.27M	From Cadene/pretrained...pytorch (log)
SE-ResNet-50	22.51	6.44	28,088,024	3,880.49M	From Cadene/pretrained...pytorch (log)
SE-ResNet-101	21.92	5.89	49,326,872	7,602.76M	From Cadene/pretrained...pytorch (log)
SE-ResNet-152	21.48	5.77	66,821,848	11,328.52M	From Cadene/pretrained...pytorch (log)
SE-ResNeXt-50 (32x4d)	21.06	5.58	27,559,896	4,258.40M	From Cadene/pretrained...pytorch (log)
SE-ResNeXt-101 (32x4d)	19.99	5.00	48,955,416	8,008.26M	From Cadene/pretrained...pytorch (log)
SENet-154	18.84	4.65	115,088,984	20,745.78M	From Cadene/pretrained...pytorch (log)
IBN-ResNet-50	23.56	6.68	25,557,032	4,110.48M	From XingangPan/IBN-Net (log)
IBN-ResNet-101	21.89	5.87	44,549,160	7,830.48M	From XingangPan/IBN-Net (log)
IBN(b)-ResNet-50	23.91	6.97	25,558,568	4,112.89M	From XingangPan/IBN-Net (log)
IBN-ResNeXt-101 (32x4d)	21.43	5.62	44,177,704	8,003.45M	From XingangPan/IBN-Net (log)
IBN-DenseNet-121	24.98	7.47	7,978,856	2,872.13M	From XingangPan/IBN-Net (log)
IBN-DenseNet-169	23.78	6.82	14,149,480	3,403.89M	From XingangPan/IBN-Net (log)
AirNet50-1x64d (r=2)	22.48	6.21	27,425,864	4,772.11M	From soeaver/AirNet-PyTorch (log)
AirNet50-1x64d (r=16)	22.91	6.46	25,714,952	4,399.97M	From soeaver/AirNet-PyTorch (log)
AirNeXt50-32x4d (r=2)	21.51	5.75	27,604,296	5,339.58M	From soeaver/AirNet-PyTorch (log)
BAM-ResNet-50	23.68	6.96	25,915,099	4,196.09M	From Jongchan/attention-module (log)
CBAM-ResNet-50	23.02	6.38	28,089,624	4,116.97M	From Jongchan/attention-module (log)
PyramidNet-101 (a=360)	22.72	6.52	42,455,070	8,743.54M	From dyhan0920/Pyramid...PyTorch (log)
DiracNetV2-18	30.61	11.17	11,511,784	1,796.62M	From szagoruyko/diracnets (log)
DiracNetV2-34	27.93	9.46	21,616,232	3,646.93M	From szagoruyko/diracnets (log)
DenseNet-121	25.11	7.80	7,978,856	2,872.13M	From dmlc/gluon-cv (log)
DenseNet-161	22.40	6.18	28,681,000	7,793.16M	From dmlc/gluon-cv (log)
DenseNet-169	23.89	6.89	14,149,480	3,403.89M	From dmlc/gluon-cv (log)
DenseNet-201	22.71	6.36	20,013,928	4,347.15M	From dmlc/gluon-cv (log)
CondenseNet-74 (C=G=4)	26.82	8.64	4,773,944	546.06M	From ShichenLiu/CondenseNet (log)
CondenseNet-74 (C=G=8)	29.76	10.49	2,935,416	291.52M	From ShichenLiu/CondenseNet (log)
PeleeNet	31.71	11.25	2,802,248	514.87M	Training (log)
WRN-50-2	22.15	6.12	68,849,128	11,405.42M	From szagoruyko/functional-zoo (log)
DRN-C-26	25.68	7.89	21,126,584	16,993.90M	From fyu/drn (log)
DRN-C-42	23.80	6.92	31,234,744	25,093.75M	From fyu/drn (log)
DRN-C-58	22.35	6.27	40,542,008	32,489.94M	From fyu/drn (log)
DRN-D-22	26.67	8.52	16,393,752	13,051.33M	From fyu/drn (log)
DRN-D-38	24.51	7.36	26,501,912	21,151.19M	From fyu/drn (log)
DRN-D-54	22.05	6.27	35,809,176	28,547.38M	From fyu/drn (log)
DRN-D-105	21.31	5.81	54,801,304	43,442.43M	From fyu/drn (log)
DPN-68	23.57	7.00	12,611,602	2,351.84M	From Cadene/pretrained...pytorch (log)
DPN-98	20.23	5.28	61,570,728	11,716.51M	From Cadene/pretrained...pytorch (log)
DPN-131	20.03	5.22	79,254,504	16,076.15M	From Cadene/pretrained...pytorch (log)
DarkNet Tiny	40.31	17.46	1,042,104	500.85M	Training (log)
DarkNet Ref	38.00	16.68	7,319,416	367.59M	Training (log)
DarkNet-53	21.44	5.56	41,609,928	7,133.86M	From dmlc/gluon-cv (log)
FishNet-150	22.85	6.38	24,959,400	6,435.02M	From kevin-ssy/FishNet (log)
SqueezeNet v1.0	38.73	17.34	1,248,424	823.67M	Training (log)
SqueezeNet v1.1	39.09	17.39	1,235,496	352.02M	Training (log)
SqueezeResNet v1.1	39.83	17.84	1,235,496	352.02M	Training (log)
1.0-SqNxt-23	42.25	18.66	724,056	287.28M	Training (log)
1.0-SqNxt-23v5	40.43	17.43	921,816	285.82M	Training (log)
ShuffleNet x0.25 (g=1)	62.00	36.77	209,746	12.35M	Training (log)
ShuffleNet x0.25 (g=3)	61.34	36.17	305,902	13.09M	Training (log)
ShuffleNet x0.5 (g=3)	43.83	20.60	718,324	41.70M	Training (log)
ShuffleNetV2 x0.5	40.61	18.30	1,366,792	43.31M	Training (log)
ShuffleNetV2 x1.0	30.94	11.23	2,278,604	149.72M	Training (log)
ShuffleNetV2 x1.5	32.38	12.37	4,406,098	320.77M	Training (log)
ShuffleNetV2 x2.0	32.04	12.10	7,601,686	595.84M	Training (log)
ShuffleNetV2b x0.5	39.81	17.82	1,366,792	43.31M	Training (log)
ShuffleNetV2b x1.0	30.39	11.01	2,279,760	150.62M	Training (log)
ShuffleNetV2c x0.5	39.87	18.11	1,366,792	43.31M	From tensorpack/tensorpack (log)
ShuffleNetV2c x1.0	30.74	11.38	2,279,760	150.62M	From tensorpack/tensorpack (log)
108-MENet-8x1 (g=3)	43.62	20.30	654,516	42.68M	Training (log)
128-MENet-8x1 (g=4)	42.10	19.13	750,796	45.98M	Training (log)
128-MENet-8x1 (g=4)	42.10	19.13	750,796	45.98M	Training (log)
160-MENet-8x1 (g=8)	43.47	20.28	850,120	45.63M	Training (log)
256-MENet-12x1 (g=4)	32.23	12.16	1,888,240	150.65M	Training (log)
348-MENet-12x1 (g=3)	31.17	11.41	3,368,128	312.00M	From clavichord93/MENet (log)
352-MENet-12x1 (g=8)	34.70	13.75	2,272,872	157.35M	From clavichord93/MENet (log)
456-MENet-24x1 (g=3)	29.57	10.43	5,304,784	567.90M	From clavichord93/MENet (log)
MobileNet x0.25	45.78	22.18	470,072	44.09M	Training (log)
MobileNet x0.5	33.94	13.30	1,331,592	155.42M	Training (log)
MobileNet x0.75	29.85	10.51	2,585,560	333.99M	Training (log)
MobileNet x1.0	26.43	8.65	4,231,976	579.80M	Training (log)
FD-MobileNet x0.25	56.19	31.38	383,160	12.95M	Training (log)
FD-MobileNet x0.5	42.62	19.69	993,928	41.84M	Training (log)
FD-MobileNet x0.75	37.91	16.01	1,833,304	86.68M	Training (log)
FD-MobileNet x1.0	33.80	13.12	2,901,288	147.46M	Training (log)
MobileNetV2 x0.25	48.08	24.12	1,516,392	34.24M	Training (log)
MobileNetV2 x0.5	35.63	14.42	1,964,736	100.13M	Training (log)
MobileNetV2 x0.75	30.82	11.26	2,627,592	198.50M	From dmlc/gluon-cv (log)
MobileNetV2 x1.0	28.51	9.90	3,504,960	329.36M	From dmlc/gluon-cv (log)
IGCV3 x0.25	53.43	28.30	1,534,020	41.29M	Training (log)
IGCV3 x0.5	39.41	17.03	1,985,528	111.12M	Training (log)
IGCV3 x1.0	28.22	9.54	3,491,688	340.79M	From homles11/IGCV3 (log)
MnasNet	31.32	11.44	4,308,816	317.67M	From zeusees/Mnasnet...Model (log)
DARTS	27.23	8.97	4,718,752	539.86M	From quark0/darts (log)
Xception	20.99	5.56	22,855,952	8,403.63M	From Cadene/pretrained...pytorch (log)
InceptionV3	21.22	5.59	23,834,568	5,743.06M	From dmlc/gluon-cv (log)
InceptionV4	20.60	5.25	42,679,816	12,304.93M	From Cadene/pretrained...pytorch (log)
InceptionResNetV2	19.96	4.94	55,843,464	13,188.64M	From Cadene/pretrained...pytorch (log)
PolyNet	19.09	4.53	95,366,600	34,821.34M	From Cadene/pretrained...pytorch (log)
NASNet-A 4@1056	25.37	7.95	5,289,978	584.90M	From Cadene/pretrained...pytorch (log)
NASNet-A 6@4032	18.17	4.24	88,753,150	23,976.44M	From Cadene/pretrained...pytorch (log)
PNASNet-5-Large	17.90	4.28	86,057,668	25,140.77M	From Cadene/pretrained...pytorch (log)

CIFAR-10

Some remarks:

• Testing subset is used for validation purpose.

Model	Error, %	Params	FLOPs/2	Remarks
ResNet-20	5.97	272,474	41.29M	Converted from GL model (log)
ResNet-56	4.52	855,770	127.06M	Converted from GL model (log)
ResNet-110	3.69	1,730,714	255.70M	Converted from GL model (log)
PreResNet-20	6.51	272,282	41.27M	Converted from GL model (log)
PreResNet-56	4.49	855,578	127.03M	Converted from GL model (log)
PreResNet-110	3.86	1,730,522	255.68M	Converted from GL model (log)
ResNeXt-29 (32x4d)	3.15	4,775,754	780.55M	Converted from GL model (log)
ResNeXt-29 (16x64d)	2.75	68,155,210	10,709.34M	From dmlc/gluon-cv (log)
WRN-16-10	2.93	17,116,634	2,414.04M	Converted from GL model (log)
WRN-28-10	2.39	36,479,194	5,246.98M	Converted from GL model (log)
WRN-40-8	2.37	35,748,314	5,176.90M	Converted from GL model (log)