index.md

Overview of OpenVINO™ Toolkit Public Models

OpenVINO™ toolkit provides a set of public models that you can use for learning and demo purposes or for developing deep learning software. Most recent version is available in the repo on Github.

The models can be downloaded via Model Downloader (<OPENVINO_INSTALL_DIR>/deployment_tools/open_model_zoo/tools/downloader).

Classification

Model Name	Implementation	OMZ Model Name	Accuracy	GFlops	mParams
AlexNet	Caffe*	alexnet		1.5	60.965
CaffeNet	Caffe*	caffenet		1.5	60.965
DenseNet 121	Caffe* TensorFlow* Caffe2*	densenet-121 densenet-121-tf densenet-121-caffe2		5.289~5.724	7.971
DenseNet 161	Caffe* TensorFlow*	densenet-161 densenet-161-tf		14.128~15.561	28.666
DenseNet 169	Caffe* TensorFlow*	densenet-169 densenet-169-tf		6.16~6.788	14.139
DenseNet 201	Caffe*	densenet-201		8.673	20.001
EfficientNet B0	TensorFlow* PyTorch*	efficientnet-b0 efficientnet-b0-pytorch	75.70/92.76 76.91/93.21	0.819	5.268
EfficientNet B0 AutoAugment	TensorFlow*	efficientnet-b0_auto_aug	76.43/93.04	0.819	5.268
EfficientNet B5	TensorFlow* PyTorch*	efficientnet-b5 efficientnet-b5-pytorch	83.33/96.67 83.69/96.71	21.252	30.303
EfficientNet B7	PyTorch*	efficientnet-b7-pytorch	84.42/96.91	77.618	66.193
EfficientNet B7 AutoAugment	TensorFlow*	efficientnet-b7_auto_aug	84.68/97.09	77.618	66.193
Inception (GoogleNet) V1	Caffe*	googlenet-v1		3.266	6.999
Inception (GoogleNet) V2	Caffe*	googlenet-v2		4.058	11.185
Inception (GoogleNet) V3	Caffe* PyTorch*	googlenet-v3 googlenet-v3-pytorch		11.469	23.817
Inception (GoogleNet) V4	Caffe*	googlenet-v4		24.584	42.648
Inception-ResNet V2	Caffe* TensorFlow*	inception-resnet-v2 inception-resnet-v2-tf		22.227~26.405	30.223~55.813
MobileNet V1 0.25 128	Caffe*	mobilenet-v1-0.25-128		0.028	0.468
MobileNet V1 0.5 160	Caffe*	mobilenet-v1-0.50-160		0.156	1.327
MobileNet V1 0.5 224	Caffe*	mobilenet-v1-0.50-224		0.304	1.327
MobileNet V1 1.0 224	Caffe* TensorFlow*	mobilenet-v1-1.0-224 mobilenet-v1-1.0-224-tf		1.148	4.221
MobileNet V2 1.0 224	Caffe* TensorFlow* PyTorch*	mobilenet-v2 mobilenet-v2-1.0-224 mobilenet-v2-pytorch		0.615~0.876	3.489
MobileNet V2 1.4 224	TensorFlow*	mobilenet-v2-1.4-224		1.183	6.087
ResNet 50	Caffe* PyTorch* Caffe2*	resnet-50 resnet-50-pytorch resnet-50-caffe2		6.996~8.216	25.53
ResNet 101	Caffe*	resnet-101		14.441	44.496
ResNet 152	Caffe*	resnet-152		21.89	60.117
SE-Inception	Caffe*	se-inception		4.091	11.922
SE-ResNet 50	Caffe*	se-resnet-50		7.775	28.061
SE-ResNet 101	Caffe*	se-resnet-101		15.239	49.274
SE-ResNet 152	Caffe*	se-resnet-152		22.709	66.746
SE-ResNeXt 50	Caffe*	se-resnext-50		8.533	27.526
SE-ResNeXt 101	Caffe*	se-resnext-101		16.054	48.886
SqueezeNet v1.0	Caffe*	squeezenet1.0		1.737	1.248
SqueezeNet v1.1	Caffe* Caffe2*	squeezenet1.1 squeezenet1.1-caffe2		0.785	1.236
VGG 16	Caffe*	vgg16		30.974	138.358
VGG 19	Caffe* Caffe2*	vgg19 vgg19-caffe2		39.3	143.667

Octave Convolutions Networks

This is are modifications of networks using Octave Convolutions. More details can be found here.

Model Name	Implementation	OMZ Model Name	Accuracy	GFlops	mParams
DenseNet 121, alpha=0.125	MXNet*	octave-densenet-121-0.125		4.883	7.977
ResNet 26, alpha=0.25	MXNet*	octave-resnet-26-0.25		3.768	15.99
ResNet 50, alpha=0.125	MXNet*	octave-resnet-50-0.125		7.221	25.551
ResNet 101, alpha=0.125	MXNet*	octave-resnet-101-0.125		13.387	44.543
ResNet 200, alpha=0.125	MXNet*	octave-resnet-200-0.125		25.407	64.667
ResNeXt 50, alpha=0.25	MXNet*	octave-resnext-50-0.25		6.444	25.02
ResNeXt 101, alpha=0.25	MXNet*	octave-resnext-101-0.25		11.521	44.169
SE-ResNet 50, alpha=0.125	MXNet*	octave-se-resnet-50-0.125		7.246	28.082

Segmentation

Semantic segmentation is an extension of object detection problem. Instead of returning bounding boxes, semantic segmentation models return a "painted" version of the input image, where the "color" of each pixel represents a certain class. These networks are much bigger than respective object detection networks, but they provide a better (pixel-level) localization of objects and they can detect areas with complex shape.

Semantic Segmentation

Model Name	Implementation	OMZ Model Name	GFlops	mParams
DeepLab V3	TensorFlow*	deeplabv3	11.469	23.819

Instance Segmentation

Instance segmentation is an extension of object detection and semantic segmentation problems. Instead of predicting a bounding box around each object instance instance segmentation model outputs pixel-wise masks for all instances.

Model Name	Implementation	OMZ Model Name	GFlops	mParams
Mask R-CNN Inception ResNet V2	TensorFlow*	mask_rcnn_inception_resnet_v2_atrous_coco	675.314	92.368
Mask R-CNN Inception V2	TensorFlow*	mask_rcnn_inception_v2_coco	54.926	21.772
Mask R-CNN ResNet 50	TensorFlow*	mask_rcnn_resnet50_atrous_coco	294.738	50.222
Mask R-CNN ResNet 101	TensorFlow*	mask_rcnn_resnet101_atrous_coco	674.58	69.188

3D Semantic Segmentation

Model Name	Implementation	OMZ Model Name	GFlops	mParams
Brain Tumor Segmentation	MXNet*	brain-tumor-segmentation-0001	409.996	38.192

Object Detection

Several detection models can be used to detect a set of the most popular objects - for example, faces, people, vehicles. Most of the networks are SSD-based and provide reasonable accuracy/performance trade-offs.

Model Name	Implementation	OMZ Model Name	GFlops	mParams
CTPN	TensorFlow*	ctpn	55.813	17.237
CenterNet (CTDET with DLAV0) 384x384	ONNX*	ctdet_coco_dlav0_384	34.994	17.911
CenterNet (CTDET with DLAV0) 512x512	ONNX*	ctdet_coco_dlav0_512	62.211	17.911
Faster R-CNN with Inception-ResNet v2	TensorFlow*	faster_rcnn_inception_resnet_v2_atrous_coco	30.687	13.307
Faster R-CNN with Inception v2	TensorFlow*	faster_rcnn_inception_v2_coco	30.687	13.307
Faster R-CNN with ResNet 50	TensorFlow*	faster_rcnn_resnet50_coco	57.203	29.162
Faster R-CNN with ResNet 101	TensorFlow*	faster_rcnn_resnet101_coco	112.052	48.128
MTCNN	Caffe*: proposal refine output	mtcnn-p mtcnn-r mtcnn-o
SSD 300	Caffe*	ssd300	62.815	26.285
SSD 512	Caffe*	ssd512	180.611	27.189
SSD with MobileNet	Caffe* TensorFlow*	mobilenet-ssd ssd_mobilenet_v1_coco	2.316~2.494	5.783~6.807
SSD with MobileNet FPN	TensorFlow*	ssd_mobilenet_v1_fpn_coco	123.309	36.188
SSD with MobileNet V2	TensorFlow*	ssd_mobilenet_v2_coco	3.775	16.818
SSD lite with MobileNet V2	TensorFlow*	ssdlite_mobilenet_v2	1.525	4.475

Face Recognition

Model Name	Implementation	OMZ Model Name	GFlops	mParams
FaceNet	TensorFlow*	facenet-20180408-102900	2.846	23.469
LResNet34E-IR,ArcFace@ms1m-refine-v1	MXNet*	face-recognition-resnet34-arcface	8.934	34.129
LResNet50E-IR,ArcFace@ms1m-refine-v1	MXNet*	face-recognition-resnet50-arcface	12.637	43.576
LResNet100E-IR,ArcFace@ms1m-refine-v2	MXNet*	face-recognition-resnet100-arcface	24.209	65.131
MobileFaceNet,ArcFace@ms1m-refine-v1	MXNet*	face-recognition-mobilefacenet-arcface	0.449	0.993
SphereFace	Caffe*	Sphereface	3.504	22.671

Human Pose Estimation

Human pose estimation task is to predict a pose: body skeleton, which consists of keypoints and connections between them, for every person in an input image or video. Keypoints are body joints, i.e. ears, eyes, nose, shoulders, knees, etc. There are two major groups of such metods: top-down and bottom-up. The first detects persons in a given frame, crops or rescales detections, then runs pose estimation network for every detection. These methods are very accurate. The second finds all keypoints in a given frame, then groups them by person instances, thus faster than previous, because network runs once.

Model Name	Implementation	OMZ Model Name	GFlops	mParams
human-pose-estimation-3d-0001	PyTorch*	human-pose-estimation-3d-0001	18.998	5.074
single-human-pose-estimation-0001	PyTorch*	single-human-pose-estimation-0001	60.125	33.165

Legal Information

[*] Other names and brands may be claimed as the property of others.