gluoncv2 0.0.58

Image classification and segmentation models for Gluon

Computer vision models on MXNet/Gluon

This is a collection of image classification, segmentation, detection, and pose estimation models. Many of them are pretrained on ImageNet-1K, CIFAR-10/100, SVHN, CUB-200-2011, Pascal VOC2012, ADE20K, Cityscapes, and COCO 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')
• SKNet ('Selective Kernel Networks')
• DIA-ResNet ('DIANet: Dense-and-Implicit Attention Network')
• PyramidNet ('Deep Pyramidal Residual Networks')
• DiracNetV2 ('DiracNets: Training Very Deep Neural Networks Without Skip-Connections')
• ShaResNet ('ShaResNet: reducing residual network parameter number by sharing weights')
• CRU-Net ('Sharing Residual Units Through Collective Tensor Factorization To Improve Deep Neural Networks')
• 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')
• Oct-ResNet ('Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution')
• Res2Net ('Res2Net: A New Multi-scale Backbone Architecture')
• WRN ('Wide Residual Networks')
• WRN-1bit ('Training wide residual networks for deployment using a single bit for each weight')
• 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')
• iSQRT-COV-ResNet ('Towards Faster Training of Global Covariance Pooling Networks by Iterative Matrix Square Root Normalization')
• i-RevNet ('i-RevNet: Deep Invertible Networks')
• BagNet ('Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet')
• DLA ('Deep Layer Aggregation')
• MSDNet ('Multi-Scale Dense Networks for Resource Efficient Image Classification')
• FishNet ('FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction')
• ESPNetv2 ('ESPNetv2: A Light-weight, Power Efficient, and General Purpose Convolutional Neural Network')
• HRNet ('Deep High-Resolution Representation Learning for Visual Recognition')
• VoVNet ('An Energy and GPU-Computation Efficient Backbone Network for Real-Time Object Detection')
• SelecSLS ('XNect: Real-time Multi-person 3D Human Pose Estimation with a Single RGB Camera')
• HarDNet ('HarDNet: A Low Memory Traffic Network')
• X-DenseNet ('Deep Expander Networks: Efficient Deep Networks from Graph Theory')
• 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')
• MobileNetV3 ('Searching for MobileNetV3')
• IGCV3 ('IGCV3: Interleaved Low-Rank Group Convolutions for Efficient Deep Neural Networks')
• GhostNet ('GhostNet: More Features from Cheap Operations')
• MnasNet ('MnasNet: Platform-Aware Neural Architecture Search for Mobile')
• DARTS ('DARTS: Differentiable Architecture Search')
• ProxylessNAS ('ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware')
• FBNet ('FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural 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')
• SPNASNet ('Single-Path NAS: Designing Hardware-Efficient ConvNets in less than 4 Hours')
• EfficientNet ('EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks')
• MixNet ('MixConv: Mixed Depthwise Convolutional Kernels')
• NIN ('Network In Network')
• RoR-3 ('Residual Networks of Residual Networks: Multilevel Residual Networks')
• RiR ('Resnet in Resnet: Generalizing Residual Architectures')
• ResDrop-ResNet ('Deep Networks with Stochastic Depth')
• Shake-Shake-ResNet ('Shake-Shake regularization')
• ShakeDrop-ResNet ('ShakeDrop Regularization for Deep Residual Learning')
• FractalNet ('FractalNet: Ultra-Deep Neural Networks without Residuals')
• NTS-Net ('Learning to Navigate for Fine-grained Classification')
• FCN-8s ('Fully Convolutional Networks for Semantic Segmentation')
• PSPNet ('Pyramid Scene Parsing Network')
• DeepLabv3 ('Rethinking Atrous Convolution for Semantic Image Segmentation')
• ICNet ('ICNet for Real-Time Semantic Segmentation on High-Resolution Images')
• SINet ('SINet: Extreme Lightweight Portrait Segmentation Networks with Spatial Squeeze Modules and Information Blocking Decoder')
• BiSeNet ('BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation')
• CenterNet ('Objects as Points')
• LFFD ('LFFD: A Light and Fast Face Detector for Edge Devices')
• AlphaPose ('RMPE: Regional Multi-person Pose Estimation')
• SimplePose ('Simple Baselines for Human Pose Estimation and Tracking')
• Lightweight OpenPose ('Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose')
• IBPPose ('Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation')
• VOCA ('Capture, Learning, and Synthesis of 3D Speaking Styles')

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 a 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 and ShuffleNetV2b are different implementations of the same architecture.
• ResNet(A) is an average downsampled ResNet intended for use as an feature extractor in some pose estimation networks.
• ResNet(D) is a dilated ResNet intended for use as an feature extractor in some segmentation networks.
• Models with *-suffix use non-standard preprocessing (see the training log).

Model	Top1	Top5	Params	FLOPs/2	Remarks
AlexNet	40.46	17.70	62,378,344	1,132.33M	Training (log)
AlexNet-b	41.08	18.53	61,100,840	714.83M	Training (log)
ZFNet	39.21	16.78	62,357,608	1,170.33M	Training (log)
ZFNet-b	35.81	14.59	107,627,624	2,479.13M	Training (log)
VGG-11	29.59	10.16	132,863,336	7,615.87M	Training (log)
VGG-13	28.37	9.50	133,047,848	11,317.65M	Training (log)
VGG-16	26.61	8.32	138,357,544	15,480.10M	Training (log)
VGG-19	25.58	7.67	143,667,240	19,642.55M	Training (log)
BN-VGG-11	28.56	9.34	132,866,088	7,630.21M	Training (log)
BN-VGG-13	27.68	8.87	133,050,792	11,341.62M	Training (log)
BN-VGG-16	25.50	7.57	138,361,768	15,506.38M	Training (log)
BN-VGG-19	23.91	6.89	143,672,744	19,671.15M	Training (log)
BN-VGG-11b	29.24	9.75	132,868,840	7,630.72M	Training (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	25.12	7.54	11,295,240	2,048.06M	Training (log)
ResNet-10	34.61	13.85	5,418,792	894.04M	Training (log)
ResNet-12	33.42	13.03	5,492,776	1,126.25M	Training (log)
ResNet-14	32.18	12.20	5,788,200	1,357.94M	Training (log)
ResNet-BC-14b	30.26	11.16	10,064,936	1,479.12M	Training (log)
ResNet-16	30.24	10.88	6,968,872	1,589.34M	Training (log)
ResNet-18 x0.25	39.31	17.40	3,937,400	270.94M	Training (log)
ResNet-18 x0.5	33.41	12.84	5,804,296	608.70M	Training (log)
ResNet-18 x0.75	30.00	10.66	8,476,056	1,129.45M	Training (log)
ResNet-18	28.09	9.51	11,689,512	1,820.41M	Training (log)
ResNet-26	26.14	8.37	17,960,232	2,746.79M	Training (log)
ResNet-BC-26b	24.86	7.58	15,995,176	2,356.67M	Training (log)
ResNet-34	24.53	7.43	21,797,672	3,672.68M	Training (log)
ResNet-BC-38b	23.50	6.72	21,925,416	3,234.21M	Training (log)
ResNet-50	22.15	6.04	25,557,032	3,877.95M	Training (log)
ResNet-50b	22.06	6.11	25,557,032	4,110.48M	Training (log)
ResNet-101	21.66	5.99	44,549,160	7,597.95M	From dmlc/gluon-cv (log)
ResNet-101b	20.26	5.12	44,549,160	7,830.48M	Training (log)
ResNet-152	20.76	5.35	60,192,808	11,321.85M	From dmlc/gluon-cv (log)
ResNet-152b	19.63	4.80	60,192,808	11,554.38M	Training (log)
PreResNet-10	34.65	14.01	5,417,128	894.19M	Training (log)
PreResNet-12	33.57	13.21	5,491,112	1,126.40M	Training (log)
PreResNet-14	32.29	12.18	5,786,536	1,358.09M	Training (log)
PreResNet-BC-14b	30.67	11.51	10,057,384	1,476.62M	Training (log)
PreResNet-16	30.21	10.81	6,967,208	1,589.49M	Training (log)
PreResNet-18 x0.25	39.62	17.78	3,935,960	270.93M	Training (log)
PreResNet-18 x0.5	33.67	13.19	5,802,440	608.73M	Training (log)
PreResNet-18 x0.75	29.96	10.68	8,473,784	1,129.51M	Training (log)
PreResNet-18	28.16	9.51	11,687,848	1,820.56M	Training (log)
PreResNet-26	26.03	8.34	17,958,568	2,746.94M	Training (log)
PreResNet-BC-26b	25.21	7.86	15,987,624	2,354.16M	Training (log)
PreResNet-34	24.55	7.51	21,796,008	3,672.83M	Training (log)
PreResNet-BC-38b	22.67	6.33	21,917,864	3,231.70M	Training (log)
PreResNet-50	22.27	6.20	25,549,480	3,875.44M	Training (log)
PreResNet-50b	22.36	6.32	25,549,480	4,107.97M	Training (log)
PreResNet-101	21.45	5.75	44,541,608	7,595.44M	From dmlc/gluon-cv (log)
PreResNet-101b	20.85	5.40	44,541,608	7,827.97M	Training (log)
PreResNet-152	20.70	5.32	60,185,256	11,319.34M	From dmlc/gluon-cv (log)
PreResNet-152b	19.90	5.00	60,185,256	11,551.87M	Training (log)
PreResNet-200b	21.10	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)	31.66	12.23	7,127,336	1,045.77M	Training (log)
ResNeXt-14 (32x2d)	32.16	12.47	7,029,416	1,031.32M	Training (log)
ResNeXt-14 (32x4d)	29.95	11.10	9,411,880	1,603.46M	Training (log)
ResNeXt-26 (32x2d)	26.34	8.50	9,924,136	1,461.06M	Training (log)
ResNeXt-26 (32x4d)	23.93	7.21	15,389,480	2,488.07M	Training (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	894.08M	Training (log)
SE-ResNet-18	27.95	9.20	11,778,592	1,820.51M	Training (log)
SE-ResNet-26	25.42	8.03	18,093,852	2,746.93M	Training (log)
SE-ResNet-BC-26b	23.44	6.82	17,395,976	2,358.07M	Training (log)
SE-ResNet-BC-38b	21.44	5.75	24,026,616	3,236.32M	Training (log)
SE-ResNet-50	21.07	5.60	28,088,024	3,880.49M	Training (log)
SE-ResNet-50b	20.58	5.33	28,088,024	4,113.02M	Training (log)
SE-ResNet-101	21.92	5.89	49,326,872	7,602.76M	From Cadene/pretrained...pytorch (log)
SE-ResNet-101b	19.46	4.62	49,326,872	7,835.29M	Training (log)
SE-ResNet-152	21.48	5.77	66,821,848	11,328.52M	From Cadene/pretrained...pytorch (log)
SE-PreResNet-10	33.60	13.06	5,461,668	894.23M	Training (log)
SE-PreResNet-18	27.67	9.38	11,776,928	1,820.66M	Training (log)
SE-PreResNet-BC-26b	22.95	6.36	17,388,424	2,355.57M	Training (log)
SE-PreResNet-BC-38b	21.42	5.63	24,019,064	3,233.81M	Training (log)
SE-PreResNet-50b	20.67	5.32	28,080,472	4,110.51M	Training (log)
SE-ResNeXt-50 (32x4d)	20.03	5.05	27,559,896	4,258.40M	From dmlc/gluon-cv (log)
SE-ResNeXt-101 (32x4d)	19.07	4.60	48,955,416	8,008.26M	From dmlc/gluon-cv (log)
SE-ResNeXt-101 (64x4d)	18.98	4.66	88,232,984	15,505.08M	From dmlc/gluon-cv (log)
SENet-16	25.34	8.06	31,366,168	5,080.55M	Training (log)
SENet-28	21.68	5.91	36,453,768	5,731.20M	Training (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)
CRU-Net-56	25.72	8.25	25,609,384	5,660.66M	From cypw/CRU-Net (log)
DenseNet-121	23.25	6.85	7,978,856	2,872.13M	Training (log)
DenseNet-161	21.82	5.92	28,681,000	7,793.16M	Training (log)
DenseNet-169	22.10	6.05	14,149,480	3,403.89M	Training (log)
DenseNet-201	21.56	5.90	20,013,928	4,347.15M	Training (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	22.87	6.58	12,611,602	2,351.84M	Training (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)
i-RevNet-301	26.97	8.97	125,120,356	14,453.87M	From jhjacobsen/pytorch-i-revnet (log)
BagNet-9	59.57	35.44	15,688,744	16,049.19M	From wielandbrendel/bag...models (log)
BagNet-17	44.76	21.52	16,213,032	15,768.77M	From wielandbrendel/bag...models (log)
BagNet-33	36.43	14.95	18,310,184	16,371.52M	From wielandbrendel/bag...models (log)
DLA-34	26.14	8.21	15,742,104	3,071.37M	From ucbdrive/dla (log)
DLA-46-C	33.84	12.86	1,301,400	585.45M	Training (log)
DLA-X-46-C	32.96	12.25	1,068,440	546.72M	Training (log)
DLA-60	23.84	7.08	22,036,632	4,255.49M	From ucbdrive/dla (log)
DLA-X-60	22.48	6.21	17,352,344	3,543.68M	From ucbdrive/dla (log)
DLA-X-60-C	30.67	10.74	1,319,832	596.06M	Training (log)
DLA-102	22.87	6.44	33,268,888	7,190.95M	From ucbdrive/dla (log)
DLA-X-102	21.97	6.02	26,309,272	5,884.94M	From ucbdrive/dla (log)
DLA-X2-102	21.12	5.53	41,282,200	9,340.61M	From ucbdrive/dla (log)
DLA-169	21.95	5.87	53,389,720	11,593.20M	From ucbdrive/dla (log)
FishNet-150	22.85	6.38	24,959,400	6,435.02M	From kevin-ssy/FishNet (log)
ESPNetv2 x0.5	43.61	21.07	1,241,332	35.36M	From sacmehta/ESPNetv2 (log)
ESPNetv2 x1.0	35.33	14.27	1,670,072	98.09M	From sacmehta/ESPNetv2 (log)
ESPNetv2 x1.25	33.14	12.73	1,965,440	138.18M	From sacmehta/ESPNetv2 (log)
ESPNetv2 x1.5	32.04	11.94	2,314,856	185.77M	From sacmehta/ESPNetv2 (log)
ESPNetv2 x2.0	28.91	9.94	3,498,136	306.93M	From sacmehta/ESPNetv2 (log)
HRNet-W18 Small V1	28.46	9.75	13,187,464	1,614.87M	From HRNet/HRNet...ation (log)
HRNet-W18 Small V2	25.75	8.02	15,597,464	2,618.54M	From HRNet/HRNet...ation (log)
HRNetV2-W18	24.01	6.85	21,299,004	4,322.66M	From HRNet/HRNet...ation (log)
HRNetV2-W30	22.31	6.07	37,712,220	8,156.14M	From HRNet/HRNet...ation (log)
HRNetV2-W32	22.27	6.07	41,232,680	8,973.31M	From HRNet/HRNet...ation (log)
HRNetV2-W40	21.72	5.71	57,557,160	12,751.34M	From HRNet/HRNet...ation (log)
HRNetV2-W44	21.73	5.92	67,064,984	14,945.95M	From HRNet/HRNet...ation (log)
HRNetV2-W48	21.41	5.78	77,469,864	17,344.29M	From HRNet/HRNet...ation (log)
HRNetV2-W64	21.08	5.52	128,059,944	28,974.95M	From HRNet/HRNet...ation (log)
VoVNet-39	23.90	6.89	22,600,296	7,086.16M	From stigma0617/VoVNet.pytorch (log)
VoVNet-57	22.95	6.60	36,640,296	8,943.09M	From stigma0617/VoVNet.pytorch (log)
SelecSLS-42b	23.31	6.78	32,458,248	2,980.62M	From rwightman/pyt...models (log)
SelecSLS-60	22.50	6.33	30,670,768	3,591.78M	From rwightman/pyt...models (log)
SelecSLS-60b	21.90	6.00	32,774,064	3,629.14M	From rwightman/pyt...models (log)
HarDNet-39DS	28.71	10.04	3,488,228	437.52M	From PingoLH/Pytorch-HarDNet (log)
HarDNet-68DS	26.43	8.47	4,180,602	788.86M	From PingoLH/Pytorch-HarDNet (log)
HarDNet-68	24.58	7.36	17,565,348	4,256.32M	From PingoLH/Pytorch-HarDNet (log)
HarDNet-85	22.61	6.42	36,670,212	9,088.58M	From PingoLH/Pytorch-HarDNet (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 v1.0	39.32	17.67	1,248,424	823.67M	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)
1.5-SqNxt-23	34.46	13.21	1,511,824	552.39M	Training (log)
1.5-SqNxt-23v5	33.48	12.68	1,953,616	550.97M	Training (log)
2.0-SqNxt-23	30.24	10.63	2,583,752	898.48M	Training (log)
2.0-SqNxt-23v5	29.27	10.24	3,366,344	897.60M	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=1)	46.22	22.38	534,484	41.16M	Training (log)
ShuffleNet x0.5 (g=3)	43.83	20.60	718,324	41.70M	Training (log)
ShuffleNet x0.75 (g=1)	39.25	16.75	975,214	86.42M	Training (log)
ShuffleNet x0.75 (g=3)	37.81	16.09	1,238,266	85.82M	Training (log)
ShuffleNet x1.0 (g=1)	34.41	13.50	1,531,936	148.13M	Training (log)
ShuffleNet x1.0 (g=2)	33.98	13.32	1,733,848	147.60M	Training (log)
ShuffleNet x1.0 (g=3)	33.96	13.29	1,865,728	145.46M	Training (log)
ShuffleNet x1.0 (g=4)	33.84	13.10	1,968,344	143.33M	Training (log)
ShuffleNet x1.0 (g=8)	33.65	13.19	2,434,768	150.76M	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	27.17	9.13	4,406,098	320.77M	Training (log)
ShuffleNetV2 x2.0	25.80	8.23	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)
ShuffleNetV2b x1.5	26.90	8.79	4,410,194	323.98M	Training (log)
ShuffleNetV2b x2.0	25.20	8.10	7,611,290	603.37M	Training (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)	27.85	9.36	3,368,128	312.00M	Training (log)
352-MENet-12x1 (g=8)	31.30	11.67	2,272,872	157.35M	Training (log)
456-MENet-24x1 (g=3)	25.02	7.80	5,304,784	567.90M	Training (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	55.44	30.53	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	29.78	10.44	2,627,592	198.50M	Training (log)
MobileNetV2 x1.0	26.77	8.64	3,504,960	329.36M	Training (log)
MobileNetV2b x0.25	48.23	25.10	1,516,312	33.18M	From dmlc/gluon-cv (log)
MobileNetV2b x0.5	35.56	14.69	1,964,448	96.42M	From dmlc/gluon-cv (log)
MobileNetV2b x0.75	30.62	11.50	2,626,968	190.52M	From dmlc/gluon-cv (log)
MobileNetV2b x1.0	27.95	9.43	3,503,872	315.51M	From dmlc/gluon-cv (log)
MobileNetV3 L/224/1.0	24.63	7.69	5,481,752	226.80M	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 x0.75	30.71	10.96	2,638,084	210.95M	Training (log)
IGCV3 x1.0	27.73	9.00	3,491,688	340.79M	Training (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	325.77M	From rwightman/pyt...models (log)
DARTS	27.23	8.97	4,718,752	539.86M	From quark0/darts (log)
ProxylessNAS CPU	24.78	7.50	4,361,648	459.96M	Training (log)
ProxylessNAS GPU	24.67	7.24	7,119,848	476.08M	Training (log)
ProxylessNAS Mobile	25.31	7.80	4,080,512	332.46M	Training (log)
ProxylessNAS Mob-14	22.96	6.51	6,857,568	597.10M	Training (log)
FBNet-Cb	25.47	7.86	5,572,200	399.26M	From rwightman/pyt...models (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)
SPNASNet	26.92	8.67	4,421,616	346.73M	From rwightman/pyt...models (log)
EfficientNet-B0	24.50	7.22	5,288,548	413.13M	Training (log)
EfficientNet-B1	22.89	6.26	7,794,184	730.44M	Training (log)
EfficientNet-B0b	22.96	6.70	5,288,548	413.13M	From rwightman/pyt...models (log)
EfficientNet-B1b	20.98	5.65	7,794,184	730.44M	From rwightman/pyt...models (log)
EfficientNet-B2b	19.94	5.16	9,109,994	1,049.29M	From rwightman/pyt...models (log)
EfficientNet-B3b	18.60	4.31	12,233,232	1,923.98M	From rwightman/pyt...models (log)
EfficientNet-B4b	17.25	3.76	19,341,616	4,597.56M	From rwightman/pyt...models (log)
EfficientNet-B5b	16.39	3.34	30,389,784	10,674.67M	From rwightman/pyt...models (log)
EfficientNet-B6b	15.96	3.12	43,040,704	19,761.35M	From rwightman/pyt...models (log)
EfficientNet-B7b	15.70	3.11	66,347,960	38,949.07M	From rwightman/pyt...models (log)
EfficientNet-B0c*	22.52	6.46	5,288,548	413.13M	From rwightman/pyt...models (log)
EfficientNet-B1c*	20.50	5.55	7,794,184	730.44M	From rwightman/pyt...models (log)
EfficientNet-B2c*	19.60	4.89	9,109,994	1,049.29M	From rwightman/pyt...models (log)
EfficientNet-B3c*	18.19	4.34	12,233,232	1,923.98M	From rwightman/pyt...models (log)
EfficientNet-B4c*	16.74	3.59	19,341,616	4,597.56M	From rwightman/pyt...models (log)
EfficientNet-B5c*	15.79	3.02	30,389,784	10,674.67M	From rwightman/pyt...models (log)
EfficientNet-B6c*	15.29	2.85	43,040,704	19,761.35M	From rwightman/pyt...models (log)
EfficientNet-B7c*	14.87	2.77	66,347,960	38,949.07M	From rwightman/pyt...models (log)
EfficientNet-B8c*	14.61	2.70	87,413,142	64,446.06M	From rwightman/pyt...models (log)
EfficientNet-Edge-Small-b*	22.48	6.29	5,438,392	2,378.09M	From rwightman/pyt...models (log)
EfficientNet-Edge-Medium-b*	21.08	5.53	6,899,496	3,700.08M	From rwightman/pyt...models (log)
EfficientNet-Edge-Large-b*	19.50	4.77	10,589,712	9,747.58M	From rwightman/pyt...models (log)
MixNet-S	24.32	7.39	4,134,606	260.26M	From rwightman/pyt...models (log)
MixNet-M	23.31	6.78	5,014,382	366.05M	From rwightman/pyt...models (log)
MixNet-L	21.53	6.03	7,329,252	590.45M	From rwightman/pyt...models (log)
ResNet(A)-50b	20.82	5.41	25,576,264	4,352.88M	From dmlc/gluon-cv (log)
ResNet(A)-101b	19.46	4.87	44,568,392	8,072.88M	From dmlc/gluon-cv (log)
ResNet(A)-152b	19.38	4.65	60,212,040	11,796.78M	From dmlc/gluon-cv (log)
ResNet(D)-50b	20.79	5.49	25,680,808	20,496.80M	From dmlc/gluon-cv (log)
ResNet(D)-101b	19.49	4.61	44,672,936	35,391.85M	From dmlc/gluon-cv (log)
ResNet(D)-152b	19.39	4.67	60,316,584	47,661.38M	From dmlc/gluon-cv (log)

CIFAR-10

Some remarks:

• Testing subset is used for validation purpose.
• Features means feature extractor output size.

Model	Error, %	Features	Params	FLOPs/2	Remarks
NIN	7.43	192	966,986	222.97M	Training (log)
ResNet-20	5.97	64	272,474	41.29M	Training (log)
ResNet-56	4.52	64	855,770	127.06M	Training (log)
ResNet-110	3.69	64	1,730,714	255.70M	Training (log)
ResNet-164(BN)	3.68	256	1,704,154	255.31M	Training (log)
ResNet-272(BN)	3.33	256	2,816,986	420.61M	Training (log)
ResNet-542(BN)	3.43	256	5,599,066	833.87M	Training (log)
ResNet-1001	3.28	256	10,328,602	1,536.40M	Training (log)
ResNet-1202	3.53	64	19,424,026	2,857.17M	Training (log)
PreResNet-20	6.51	64	272,282	41.27M	Training (log)
PreResNet-56	4.49	64	855,578	127.03M	Training (log)
PreResNet-110	3.86	64	1,730,522	255.68M	Training (log)
PreResNet-164(BN)	3.64	256	1,703,258	255.08M	Training (log)
PreResNet-272(BN)	3.25	256	2,816,090	420.38M	Training (log)
PreResNet-542(BN)	3.14	256	5,598,170	833.64M	Training (log)
PreResNet-1001	2.65	256	10,327,706	1,536.18M	Training (log)
PreResNet-1202	3.39	64	19,423,834	2,857.14M	Training (log)
ResNeXt-20 (1x64d)	4.33	1024	3,446,602	538.36M	Training (log)
ResNeXt-20 (2x32d)	4.53	1024	2,672,458	425.15M	Training (log)
ResNeXt-20 (4x16d)	4.70	1024	2,285,386	368.55M	Training (log)
ResNeXt-20 (8x8d)	4.66	1024	2,091,850	340.25M	Training (log)
ResNeXt-20 (16x4d)	4.04	1024	1,995,082	326.10M	Training (log)
ResNeXt-20 (32x2d)	4.61	1024	1,946,698	319.03M	Training (log)
ResNeXt-20 (64x1d)	4.93	1024	1,922,506	315.49M	Training (log)
ResNeXt-20 (2x64d)	4.03	1024	6,198,602	987.98M	Training (log)
ResNeXt-20 (4x32d)	3.73	1024	4,650,314	761.57M	Training (log)
ResNeXt-20 (8x16d)	4.04	1024	3,876,170	648.37M	Training (log)
ResNeXt-20 (16x8d)	3.94	1024	3,489,098	591.77M	Training (log)
ResNeXt-20 (32x4d)	4.20	1024	3,295,562	563.47M	Training (log)
ResNeXt-20 (64x2d)	4.38	1024	3,198,794	549.32M	Training (log)
ResNeXt-56 (1x64d)	2.87	1024	9,317,194	1,399.33M	Training (log)
ResNeXt-56 (2x32d)	3.01	1024	6,994,762	1,059.72M	Training (log)
ResNeXt-56 (4x16d)	3.11	1024	5,833,546	889.91M	Training (log)
ResNeXt-56 (8x8d)	3.07	1024	5,252,938	805.01M	Training (log)
ResNeXt-56 (16x4d)	3.12	1024	4,962,634	762.56M	Training (log)
ResNeXt-56 (32x2d)	3.14	1024	4,817,482	741.34M	Training (log)
ResNeXt-56 (64x1d)	3.41	1024	4,744,906	730.72M	Training (log)
ResNeXt-29 (32x4d)	3.15	1024	4,775,754	780.55M	Training (log)
ResNeXt-29 (16x64d)	2.41	1024	68,155,210	10,709.34M	Training (log)
ResNeXt-272 (1x64d)	2.55	1024	44,540,746	6,565.15M	Training (log)
ResNeXt-272 (2x32d)	2.74	1024	32,928,586	4,867.11M	Training (log)
SE-ResNet-20	6.01	64	274,847	41.30M	Training (log)
SE-ResNet-56	4.13	64	862,889	127.07M	Training (log)
SE-ResNet-110	3.63	64	1,744,952	255.72M	Training (log)
SE-ResNet-164(BN)	3.39	256	1,906,258	255.52M	Training (log)
SE-ResNet-272(BN)	3.39	256	3,153,826	420.96M	Training (log)
SE-ResNet-542(BN)	3.47	256	6,272,746	834.57M	Training (log)
SE-PreResNet-20	6.18	64	274,559	41.30M	Training (log)
SE-PreResNet-56	4.51	64	862,601	127.07M	Training (log)
SE-PreResNet-110	4.54	64	1,744,664	255.72M	Training (log)
SE-PreResNet-164(BN)	3.73	256	1,904,882	255.29M	Training (log)
SE-PreResNet-272(BN)	3.39	256	3,152,450	420.73M	Training (log)
SE-PreResNet-542(BN)	3.08	256	6,271,370	834.34M	Training (log)
PyramidNet-110 (a=48)	3.72	64	1,772,706	408.37M	Training (log)
PyramidNet-110 (a=84)	2.98	100	3,904,446	778.15M	Training (log)
PyramidNet-110 (a=270)	2.51	286	28,485,477	4,730.60M	Training (log)
PyramidNet-164 (a=270, BN)	2.42	1144	27,216,021	4,608.81M	Training (log)
PyramidNet-200 (a=240, BN)	2.44	1024	26,752,702	4,563.40M	Training (log)
PyramidNet-236 (a=220, BN)	2.47	944	26,969,046	4,631.32M	Training (log)
PyramidNet-272 (a=200, BN)	2.39	864	26,210,842	4,541.36M	Training (log)
DenseNet-40 (k=12)	5.61	258	599,050	210.80M	Training (log)
DenseNet-BC-40 (k=12)	6.43	132	176,122	74.89M	Training (log)
DenseNet-BC-40 (k=24)	4.52	264	690,346	293.09M	Training (log)
DenseNet-BC-40 (k=36)	4.04	396	1,542,682	654.60M	Training (log)
DenseNet-100 (k=12)	3.66	678	4,068,490	1,353.55M	Training (log)
DenseNet-100 (k=24)	3.13	1356	16,114,138	5,354.19M	Training (log)
DenseNet-BC-100 (k=12)	4.16	342	769,162	298.45M	Training (log)
DenseNet-BC-190 (k=40)	2.52	2190	25,624,430	9,400.45M	Training (log)
DenseNet-BC-250 (k=24)	2.67	1734	15,324,406	5,519.54M	Training (log)
X-DenseNet-BC-40-2 (k=24)	5.31	264	690,346	293.09M	Training (log)
X-DenseNet-BC-40-2 (k=36)	4.37	396	1,542,682	654.60M	Training (log)
WRN-16-10	2.93	640	17,116,634	2,414.04M	Training (log)
WRN-28-10	2.39	640	36,479,194	5,246.98M	Training (log)
WRN-40-8	2.37	512	35,748,314	5,176.90M	Training (log)
WRN-20-10-1bit	3.26	640	26,737,140	4,019.14M	Training (log)
WRN-20-10-32bit	3.14	640	26,737,140	4,019.14M	Training (log)
RoR-3-56	5.43	64	762,746	113.43M	Training (log)
RoR-3-110	4.35	64	1,637,690	242.07M	Training (log)
RoR-3-164	3.93	64	2,512,634	370.72M	Training (log)
RiR	3.28	384	9,492,980	1,281.08M	Training (log)
Shake-Shake-ResNet-20-2x16d	5.15	64	541,082	81.78M	Training (log)
Shake-Shake-ResNet-26-2x32d	3.17	64	2,923,162	428.89M	Training (log)
DIA-ResNet-20	6.22	64	286,866	41.34M	Training (log)
DIA-ResNet-56	5.05	64	870,162	127.18M	Training (log)
DIA-ResNet-110	4.10	64	1,745,106	255.94M	Training (log)
DIA-ResNet-164(BN)	3.50	256	1,923,002	259.18M	Training (log)
DIA-PreResNet-20	6.42	64	286,674	41.31M	Training (log)
DIA-PreResNet-56	4.83	64	869,970	127.15M	Training (log)
DIA-PreResNet-110	4.25	64	1,744,914	255.92M	Training (log)
DIA-PreResNet-164(BN)	3.56	256	1,922,106	258.95M	Training (log)

CIFAR-100

Some remarks:

• Testing subset is used for validation purpose.

Model	Error, %	Params	FLOPs/2	Remarks
NIN	28.39	984,356	224.08M	Training (log)
ResNet-20	29.64	278,324	41.30M	Training (log)
ResNet-56	24.88	861,620	127.06M	Training (log)
ResNet-110	22.80	1,736,564	255.71M	Training (log)
ResNet-164(BN)	20.44	1,727,284	255.33M	Training (log)
ResNet-272(BN)	20.07	2,840,116	420.63M	Training (log)
ResNet-542(BN)	19.32	5,622,196	833.89M	Training (log)
ResNet-1001	19.79	10,351,732	1,536.43M	Training (log)
ResNet-1202	21.56	19,429,876	2,857.17M	Training (log)
PreResNet-20	30.22	278,132	41.28M	Training (log)
PreResNet-56	25.05	861,428	127.04M	Training (log)
PreResNet-110	22.67	1,736,372	255.68M	Training (log)
PreResNet-164(BN)	20.18	1,726,388	255.10M	Training (log)
PreResNet-272(BN)	19.63	2,839,220	420.40M	Training (log)
PreResNet-542(BN)	18.71	5,621,300	833.66M	Training (log)
PreResNet-1001	18.41	10,350,836	1,536.20M	Training (log)
ResNeXt-20 (1x64d)	21.97	3,538,852	538.45M	Training (log)
ResNeXt-20 (2x32d)	22.55	2,764,708	425.25M	Training (log)
ResNeXt-20 (4x16d)	23.04	2,377,636	368.65M	Training (log)
ResNeXt-20 (8x8d)	22.82	2,184,100	340.34M	Training (log)
ResNeXt-20 (16x4d)	22.82	2,087,332	326.19M	Training (log)
ResNeXt-20 (32x2d)	21.73	2,038,948	319.12M	Training (log)
ResNeXt-20 (64x1d)	23.53	2,014,756	315.58M	Training (log)
ResNeXt-20 (2x64d)	20.60	6,290,852	988.07M	Training (log)
ResNeXt-20 (4x32d)	21.31	4,742,564	761.66M	Training (log)
ResNeXt-20 (8x16d)	21.72	3,968,420	648.46M	Training (log)
ResNeXt-20 (16x8d)	21.73	3,581,348	591.86M	Training (log)
ResNeXt-20 (32x4d)	22.13	3,387,812	563.56M	Training (log)
ResNeXt-20 (64x2d)	22.35	3,291,044	549.41M	Training (log)
ResNeXt-56 (1x64d)	18.25	9,409,444	1,399.42M	Training (log)
ResNeXt-56 (2x32d)	17.86	7,087,012	1,059.81M	Training (log)
ResNeXt-56 (4x16d)	18.09	5,925,796	890.01M	Training (log)
ResNeXt-56 (8x8d)	18.06	5,345,188	805.10M	Training (log)
ResNeXt-56 (16x4d)	18.24	5,054,884	762.65M	Training (log)
ResNeXt-56 (32x2d)	18.60	4,909,732	741.43M	Training (log)
ResNeXt-56 (64x1d)	18.16	4,837,156	730.81M	Training (log)
ResNeXt-29 (32x4d)	19.50	4,868,004	780.64M	Training (log)
ResNeXt-29 (16x64d)	16.93	68,247,460	10,709.43M	Training (log)
ResNeXt-272 (1x64d)	19.11	44,632,996	6,565.25M	Training (log)
ResNeXt-272 (2x32d)	18.34	33,020,836	4,867.20M	Training (log)
SE-ResNet-20	28.54	280,697	41.30M	Training (log)
SE-ResNet-56	22.94	868,739	127.07M	Training (log)
SE-ResNet-110	20.86	1,750,802	255.72M	Training (log)
SE-ResNet-164(BN)	19.95	1,929,388	255.54M	Training (log)
SE-ResNet-272(BN)	19.07	3,176,956	420.98M	Training (log)
SE-ResNet-542(BN)	18.87	6,295,876	834.59M	Training (log)
SE-PreResNet-20	28.31	280,409	41.31M	Training (log)
SE-PreResNet-56	23.05	868,451	127.08M	Training (log)
SE-PreResNet-110	22.61	1,750,514	255.73M	Training (log)
SE-PreResNet-164(BN)	20.05	1,928,012	255.31M	Training (log)
SE-PreResNet-272(BN)	19.13	3,175,580	420.75M	Training (log)
SE-PreResNet-542(BN)	19.45	6,294,500	834.36M	Training (log)
PyramidNet-110 (a=48)	20.95	1,778,556	408.38M	Training (log)
PyramidNet-110 (a=84)	18.87	3,913,536	778.16M	Training (log)
PyramidNet-110 (a=270)	17.10	28,511,307	4,730.62M	Training (log)
PyramidNet-164 (a=270, BN)	16.70	27,319,071	4,608.91M	Training (log)
PyramidNet-200 (a=240, BN)	16.09	26,844,952	4,563.49M	Training (log)
PyramidNet-236 (a=220, BN)	16.34	27,054,096	4,631.41M	Training (log)
PyramidNet-272 (a=200, BN)	16.19	26,288,692	4,541.43M	Training (log)
DenseNet-40 (k=12)	24.90	622,360	210.82M	Training (log)
DenseNet-BC-40 (k=12)	28.41	188,092	74.90M	Training (log)
DenseNet-BC-40 (k=24)	22.67	714,196	293.11M	Training (log)
DenseNet-BC-40 (k=36)	20.50	1,578,412	654.64M	Training (log)
DenseNet-100 (k=12)	19.64	4,129,600	1,353.62M	Training (log)
DenseNet-100 (k=24)	18.08	16,236,268	5,354.32M	Training (log)
DenseNet-BC-100 (k=12)	21.19	800,032	298.48M	Training (log)
DenseNet-BC-250 (k=24)	17.39	15,480,556	5,519.69M	Training (log)
X-DenseNet-BC-40-2 (k=24)	23.96	714,196	293.11M	Training (log)
X-DenseNet-BC-40-2 (k=36)	21.65	1,578,412	654.64M	Training (log)
WRN-16-10	18.95	17,174,324	2,414.09M	Training (log)
WRN-28-10	17.88	36,536,884	5,247.04M	Training (log)
WRN-40-8	18.03	35,794,484	5,176.95M	Training (log)
WRN-20-10-1bit	19.04	26,794,920	4,022.81M	Training (log)
WRN-20-10-32bit	18.12	26,794,920	4,022.81M	Training (log)
RoR-3-56	25.49	768,596	113.43M	Training (log)
RoR-3-110	23.64	1,643,540	242.08M	Training (log)
RoR-3-164	22.34	2,518,484	370.72M	Training (log)
RiR	19.23	9,527,720	1,283.29M	Training (log)
Shake-Shake-ResNet-20-2x16d	29.22	546,932	81.79M	Training (log)
Shake-Shake-ResNet-26-2x32d	18.80	2,934,772	428.90M	Training (log)
DIA-ResNet-20	27.71	292,716	41.34M	Training (log)
DIA-ResNet-56	24.35	876,012	127.18M	Training (log)
DIA-ResNet-110	22.11	1,750,956	255.95M	Training (log)
DIA-ResNet-164(BN)	19.53	1,946,132	259.20M	Training (log)
DIA-PreResNet-20	28.37	292,524	41.32M	Training (log)
DIA-PreResNet-56	25.05	875,820	127.16M	Training (log)
DIA-PreResNet-110	22.69	1,750,764	255.92M	Training (log)
DIA-PreResNet-164(BN)	19.99	1,945,236	258.97M	Training (log)

SVHN

Model	Error, %	Params	FLOPs/2	Remarks
NIN	3.76	966,986	222.97M	Training (log)
ResNet-20	3.43	272,474	41.29M	Training (log)
ResNet-56	2.75	855,770	127.06M	Training (log)
ResNet-110	2.45	1,730,714	255.70M	Training (log)
ResNet-164(BN)	2.42	1,704,154	255.31M	Training (log)
ResNet-272(BN)	2.43	2,816,986	420.61M	Training (log)
ResNet-542(BN)	2.34	5,599,066	833.87M	Training (log)
ResNet-1001	2.41	10,328,602	1,536.40M	Training (log)
PreResNet-20	3.22	272,282	41.27M	Training (log)
PreResNet-56	2.80	855,578	127.03M	Training (log)
PreResNet-110	2.79	1,730,522	255.68M	Training (log)
PreResNet-164(BN)	2.58	1,703,258	255.08M	Training (log)
PreResNet-272(BN)	2.34	2,816,090	420.38M	Training (log)
PreResNet-542(BN)	2.36	5,598,170	833.64M	Training (log)
ResNeXt-20 (1x64d)	2.98	3,446,602	538.36M	Training (log)
ResNeXt-20 (2x32d)	2.96	2,672,458	425.15M	Training (log)
ResNeXt-20 (4x16d)	3.17	2,285,386	368.55M	Training (log)
ResNeXt-20 (8x8d)	3.18	2,091,850	340.25M	Training (log)
ResNeXt-20 (16x4d)	3.21	1,995,082	326.10M	Training (log)
ResNeXt-20 (32x2d)	3.27	1,946,698	319.03M	Training (log)
ResNeXt-20 (64x1d)	3.42	1,922,506	315.49M	Training (log)
ResNeXt-20 (2x64d)	2.83	6,198,602	987.98M	Training (log)
ResNeXt-20 (4x32d)	2.98	4,650,314	761.57M	Training (log)
ResNeXt-20 (8x16d)	3.01	3,876,170	648.37M	Training (log)
ResNeXt-20 (16x8d)	2.93	3,489,098	591.77M	Training (log)
ResNeXt-20 (32x4d)	3.09	3,295,562	563.47M	Training (log)
ResNeXt-20 (64x2d)	3.14	3,198,794	549.32M	Training (log)
ResNeXt-56 (1x64d)	2.42	9,317,194	1,399.33M	Training (log)
ResNeXt-56 (2x32d)	2.46	6,994,762	1,059.72M	Training (log)
ResNeXt-56 (4x16d)	2.44	5,833,546	889.91M	Training (log)
ResNeXt-56 (8x8d)	2.47	5,252,938	805.01M	Training (log)
ResNeXt-56 (16x4d)	2.56	4,962,634	762.56M	Training (log)
ResNeXt-56 (32x2d)	2.53	4,817,482	741.34M	Training (log)
ResNeXt-56 (64x1d)	2.55	4,744,906	730.72M	Training (log)
ResNeXt-29 (32x4d)	2.80	4,775,754	780.55M	Training (log)
ResNeXt-29 (16x64d)	2.68	68,155,210	10,709.34M	Training (log)
ResNeXt-272 (1x64d)	2.35	44,540,746	6,565.15M	Training (log)
ResNeXt-272 (2x32d)	2.44	32,928,586	4,867.11M	Training (log)
SE-ResNet-20	3.23	274,847	41.30M	Training (log)
SE-ResNet-56	2.64	862,889	127.07M	Training (log)
SE-ResNet-110	2.35	1,744,952	255.72M	Training (log)
SE-ResNet-164(BN)	2.45	1,906,258	255.52M	Training (log)
SE-ResNet-272(BN)	2.38	3,153,826	420.96M	Training (log)
SE-ResNet-542(BN)	2.26	6,272,746	834.57M	Training (log)
SE-PreResNet-20	3.24	274,559	41.30M	Training (log)
SE-PreResNet-56	2.71	862,601	127.07M	Training (log)
SE-PreResNet-110	2.59	1,744,664	255.72M	Training (log)
SE-PreResNet-164(BN)	2.56	1,904,882	255.29M	Training (log)
SE-PreResNet-272(BN)	2.49	3,152,450	420.73M	Training (log)
SE-PreResNet-542(BN)	2.47	6,271,370	834.34M	Training (log)
PyramidNet-110 (a=48)	2.47	1,772,706	408.37M	Training (log)
PyramidNet-110 (a=84)	2.43	3,904,446	778.15M	Training (log)
PyramidNet-110 (a=270)	2.38	28,485,477	4,730.60M	Training (log)
PyramidNet-164 (a=270, BN)	2.33	27,216,021	4,608.81M	Training (log)
PyramidNet-200 (a=240, BN)	2.32	26,752,702	4,563.40M	Training (log)
PyramidNet-236 (a=220, BN)	2.35	26,969,046	4,631.32M	Training (log)
PyramidNet-272 (a=200, BN)	2.40	26,210,842	4,541.36M	Training (log)
DenseNet-40 (k=12)	3.05	599,050	210.80M	Training (log)
DenseNet-BC-40 (k=12)	3.20	176,122	74.89M	Training (log)
DenseNet-BC-40 (k=24)	2.90	690,346	293.09M	Training (log)
DenseNet-BC-40 (k=36)	2.60	1,542,682	654.60M	Training (log)
DenseNet-100 (k=12)	2.60	4,068,490	1,353.55M	Training (log)
X-DenseNet-BC-40-2 (k=24)	2.87	690,346	293.09M	Training (log)
X-DenseNet-BC-40-2 (k=36)	2.74	1,542,682	654.60M	Training (log)
WRN-16-10	2.78	17,116,634	2,414.04M	Training (log)
WRN-28-10	2.71	36,479,194	5,246.98M	Training (log)
WRN-40-8	2.54	35,748,314	5,176.90M	Training (log)
WRN-20-10-1bit	2.73	26,737,140	4,019.14M	Training (log)
WRN-20-10-32bit	2.59	26,737,140	4,019.14M	Training (log)
RoR-3-56	2.69	762,746	113.43M	Training (log)
RoR-3-110	2.57	1,637,690	242.07M	Training (log)
RoR-3-164	2.73	2,512,634	370.72M	Training (log)
RiR	2.68	9,492,980	1,281.08M	Training (log)
Shake-Shake-ResNet-20-2x16d	3.17	541,082	81.78M	Training (log)
Shake-Shake-ResNet-26-2x32d	2.62	2,923,162	428.89M	Training (log)
DIA-ResNet-20	3.23	286,866	41.34M	Training (log)
DIA-ResNet-56	2.68	870,162	127.18M	Training (log)
DIA-ResNet-110	2.47	1,745,106	255.94M	Training (log)
DIA-ResNet-164(BN)	2.44	1,923,002	259.18M	Training (log)
DIA-PreResNet-20	3.03	286,674	41.31M	Training (log)
DIA-PreResNet-56	2.80	869,970	127.15M	Training (log)
DIA-PreResNet-110	2.42	1,744,914	255.92M	Training (log)
DIA-PreResNet-164(BN)	2.56	1,922,106	258.95M	Training (log)

CUB-200-2011

Model	Error, %	Params	FLOPs/2	Remarks
ResNet-10	27.65	5,008,392	893.63M	Training (log)
ResNet-12	26.58	5,082,376	1,125.84M	Training (log)
ResNet-14	24.35	5,377,800	1,357.53M	Training (log)
ResNet-16	23.21	6,558,472	1,588.93M	Training (log)
ResNet-18	23.30	11,279,112	1,820.00M	Training (log)
ResNet-26	22.52	17,549,832	2,746.38M	Training (log)
SE-ResNet-10	27.39	5,052,932	893.67M	Training (log)
SE-ResNet-12	26.04	5,127,496	1,125.88M	Training (log)
SE-ResNet-14	23.63	5,425,104	1,357.58M	Training (log)
SE-ResNet-16	23.21	6,614,240	1,588.99M	Training (log)
SE-ResNet-18	23.08	11,368,192	1,820.10M	Training (log)
SE-ResNet-26	22.51	17,683,452	2,746.52M	Training (log)
MobileNet x1.0	23.46	3,411,976	578.98M	Training (log)
ProxylessNAS Mobile	21.88	3,055,712	331.44M	Training (log)
NTS-Net	13.26	28,623,333	33,361.39M	From yangze0930/NTS-Net (log)

Pascal VOC20102

Model	Extractor	Pix.Acc.,%	mIoU,%	Params	FLOPs/2	Remarks
PSPNet	ResNet(D)-101b	98.09	81.44	65,708,501	230,586.69M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-101b	97.95	80.24	58,754,773	47,624.54M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-152b	98.11	81.20	74,398,421	59,894.06M	From dmlc/gluon-cv (log)
FCN-8s(d)	ResNet(D)-101b	97.80	80.40	52,072,917	196,562.96M	From dmlc/gluon-cv (log)

ADE20K

Model	Extractor	Pix.Acc.,%	mIoU,%	Params	FLOPs/2	Remarks
PSPNet	ResNet(D)-50b	79.37	36.87	46,782,550	162,410.82M	From dmlc/gluon-cv (log)
PSPNet	ResNet(D)-101b	79.93	37.97	65,774,678	230,824.47M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-50b	79.72	37.13	39,795,798	32,755.38M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-101b	80.21	37.84	58,787,926	47,650.43M	From dmlc/gluon-cv (log)
FCN-8s(d)	ResNet(D)-50b	76.92	33.39	33,146,966	128,387.08M	From dmlc/gluon-cv (log)
FCN-8s(d)	ResNet(D)-101b	79.01	35.88	52,139,094	196,800.73M	From dmlc/gluon-cv (log)

Cityscapes

Model	Extractor	Pix.Acc.,%	mIoU,%	Params	FLOPs/2	Remarks
PSPNet	ResNet(D)-101b	96.17	71.72	65,707,475	230,583.01M	From dmlc/gluon-cv (log)
ICNet	ResNet(D)-50b	95.50	64.02	47,489,184	14,241.91M	From dmlc/gluon-cv (log)
SINet	-	93.66	60.31	119,418	1,411.97M	From clovaai/c3_sinet (log)

COCO Semantic Segmentation

Model	Extractor	Pix.Acc.,%	mIoU,%	Params	FLOPs/2	Remarks
PSPNet	ResNet(D)-101b	92.05	67.41	65,708,501	230,586.69M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-101b	92.19	67.73	58,754,773	47,624.54M	From dmlc/gluon-cv (log)
DeepLabv3	ResNet(D)-152b	92.24	68.99	74,398,421	275,084.22M	From dmlc/gluon-cv (log)
FCN-8s(d)	ResNet(D)-101b	91.44	60.11	52,072,917	196,562.96M	From dmlc/gluon-cv (log)

CelebAMask-HQ

Model	Extractor	Params	FLOPs/2	Remarks
BiSeNet	ResNet-18	13,300,416	-	From zllrunning/face...Torch (log)

COCO Keypoints Detection

Model	Extractor	OKS AP, %	Params	FLOPs/2	Remarks
AlphaPose	Fast-SE-ResNet-101b	74.15/91.59/80.68	59,569,873	9,553.15M	From dmlc/gluon-cv (log)
SimplePose	ResNet-18	66.31/89.20/73.41	15,376,721	1,799.25M	From dmlc/gluon-cv (log)
SimplePose	ResNet-50b	71.02/91.23/78.57	33,999,697	4,041.06M	From dmlc/gluon-cv (log)
SimplePose	ResNet-101b	72.44/92.18/79.76	52,991,825	7,685.04M	From dmlc/gluon-cv (log)
SimplePose	ResNet-152b	72.53/92.14/79.61	68,635,473	11,332.86M	From dmlc/gluon-cv (log)
SimplePose	ResNet(A)-50b	71.70/91.31/78.66	34,018,929	4,278.56M	From dmlc/gluon-cv (log)
SimplePose	ResNet(A)-101b	72.97/92.24/80.81	53,011,057	7,922.54M	From dmlc/gluon-cv (log)
SimplePose	ResNet(A)-152b	73.44/92.27/80.72	68,654,705	11,570.36M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	ResNet-18	66.25/89.17/74.32	12,858,208	1,960.96M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	ResNet-50b	71.10/91.28/78.67	25,582,944	4,221.30M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	1.0 MobileNet-224	64.10/88.06/71.23	5,019,744	751.36M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	1.0 MobileNetV2b-224	63.74/88.12/71.06	4,102,176	495.95M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	MobileNetV3 Small 224/1.0	54.34/83.67/59.35	2,625,088	236.51M	From dmlc/gluon-cv (log)
SimplePose(Mobile)	MobileNetV3 Large 224/1.0	63.67/88.91/70.82	4,768,336	403.97M	From dmlc/gluon-cv (log)
Lightweight OpenPose 2D	MobileNet	39.99/65.95/40.70	4,091,698	8,948.96M	From Daniil-Osokin/lighw...ch (log)
Lightweight OpenPose 3D	MobileNet	39.99/65.95/40.70	5,085,983	11,049.43M	From Daniil-Osokin/li...3d...ch (log)
IBPPose	-	64.86/83.62/70.12	95,827,784	57,193.82M	From jialee93/Improved...Parts (log)