Overview of OpenVINO™ Toolkit Public Pre-Trained Models¶
OpenVINO toolkit provides a set of public pre-trained 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 table Public Pre-Trained Models Device Support summarizes devices supported by each model.
You can download models and convert them into Inference Engine format (*.xml + *.bin) using the OpenVINO™ Model Downloader and other automation tools.
Classification¶
Model Name Model Version |
Implementation |
Top 1 Accuracy |
Top 5 Accuracy |
GFlops |
mParams |
---|---|---|---|---|---|
AlexNet alexnet |
Caffe |
56.60% |
79.81% |
1.5 |
60.965 |
AntiSpoofNet anti-spoof-mn3 |
PyTorch |
3.81% |
0.15 |
3.02 |
|
CaffeNet caffenet |
Caffe |
56.71% |
79.91% |
1.5 |
60.965 |
DenseNet 121 densenet-121 |
Caffe |
74.42% |
92.14% |
5.724 |
7.971 |
DenseNet 121 densenet-121-tf |
TensorFlow |
74.46% |
92.13% |
5.7287 |
7.9714 |
DenseNet 121 densenet-121-caffe2 |
Caffe2 |
74.90% |
92.19% |
5.723 |
7.971 |
DenseNet 161 densenet-161 |
Caffe |
77.55% |
93.92% |
15.561 |
28.666 |
DenseNet 161 densenet-161-tf |
TensorFlow |
76.45% |
93.23% |
14.128 |
28.666 |
DenseNet 169 densenet-169 |
Caffe |
76.11% |
93.11% |
6.788 |
14.139 |
DenseNet 169 densenet-169-tf |
TensorFlow |
76.14% |
93.12% |
6.7932 |
14.1389 |
DenseNet 201 densenet-201 |
Caffe |
76.89% |
93.56% |
8.673 |
20.001 |
DenseNet 201 densenet-201-tf |
TensorFlow |
76.93% |
93.56% |
8.6786 |
20.0013 |
DLA 34 dla-34 |
PyTorch |
74.64% |
92.06% |
6.1368 |
15.7344 |
EfficientNet B0 efficientnet-b0 |
TensorFlow |
75.70% |
92.76% |
0.819 |
5.268 |
EfficientNet B0 efficientnet-b0-pytorch |
PyTorch |
76.91% |
93.21% |
0.819 |
5.268 |
EfficientNet B0 AutoAugment efficientnet-b0_auto_aug |
TensorFlow |
76.43% |
93.04% |
0.819 |
5.268 |
EfficientNet B5 efficientnet-b5 |
TensorFlow |
83.33% |
96.67% |
21.252 |
30.303 |
EfficientNet B5 efficientnet-b5-pytorch |
PyTorch |
83.69% |
96.71% |
21.252 |
30.303 |
EfficientNet B7 efficientnet-b7-pytorch |
PyTorch |
84.42% |
96.91% |
77.618 |
66.193 |
EfficientNet B7 AutoAugment efficientnet-b7_auto_aug |
TensorFlow |
84.68% |
97.09% |
77.618 |
66.193 |
HBONet 1.0 hbonet-1.0 |
PyTorch |
73.10% |
91.00% |
0.6208 |
4.5443 |
HBONet 0.5 hbonet-0.5 |
PyTorch |
67.00% |
86.90% |
0.1977 |
2.5287 |
HBONet 0.25 hbonet-0.25 |
PyTorch |
57.30% |
79.80% |
0.0758 |
1.9299 |
Inception (GoogleNet) V1 googlenet-v1 |
Caffe |
68.93% |
89.14% |
3.266 |
6.999 |
Inception (GoogleNet) V1 googlenet-v1-tf |
TensorFlow |
69.81% |
89.60% |
3.016 |
6.619 |
Inception (GoogleNet) V2 googlenet-v2 |
Caffe |
72.02% |
90.84% |
4.058 |
11.185 |
Inception (GoogleNet) V2 googlenet-v2-tf |
TensorFlow |
74.09% |
91.80% |
4.058 |
11.185 |
Inception (GoogleNet) V3 googlenet-v3 |
TensorFlow |
77.90% |
93.81% |
11.469 |
23.819 |
Inception (GoogleNet) V3 googlenet-v3-pytorch |
PyTorch |
77.69% |
93.70% |
11.469 |
23.817 |
Inception (GoogleNet) V4 googlenet-v4-tf |
TensorFlow |
80.20% |
95.21% |
24.584 |
42.648 |
Inception-ResNet V2 inception-resnet-v2-tf |
TensorFlow |
80.14% |
95.10% |
22.227 |
30.223 |
MixNet L mixnet-l |
TensorFlow |
78.30% |
93.91% |
0.565 |
7.3 |
MobileNet V1 0.25 128 mobilenet-v1-0.25-128 |
Caffe |
40.54% |
65.00% |
0.028 |
0.468 |
MobileNet V1 0.5 160 mobilenet-v1-0.50-160 |
Caffe |
59.86% |
82.04% |
0.156 |
1.327 |
MobileNet V1 0.5 224 mobilenet-v1-0.50-224 |
Caffe |
63.04% |
84.93% |
0.304 |
1.327 |
MobileNet V1 1.0 224 mobilenet-v1-1.0-224 |
Caffe |
69.50% |
89.22% |
1.148 |
4.221 |
MobileNet V1 1.0 224 mobilenet-v1-1.0-224-tf |
TensorFlow |
71.03% |
89.94% |
1.148 |
4.221 |
MobileNet V2 1.0 224 mobilenet-v2 |
Caffe |
71.22% |
90.18% |
0.876 |
3.489 |
MobileNet V2 1.0 224 mobilenet-v2-1.0-224 |
TensorFlow |
71.85% |
90.69% |
0.615 |
3.489 |
MobileNet V2 1.0 224 mobilenet-v2-pytorch |
PyTorch |
71.90% |
90.30% |
0.615 |
3.489 |
MobileNet V2 1.4 224 mobilenet-v2-1.4-224 |
TensorFlow |
74.09% |
91.97% |
1.183 |
6.087 |
MobileNet V3 Small 1.0 mobilenet-v3-small-1.0-224-tf |
TensorFlow |
67.36% |
87.45% |
0.121 |
2.537 |
MobileNet V3 Large 1.0 mobilenet-v3-large-1.0-224-tf |
TensorFlow |
75.70% |
92.76% |
0.4536 |
5.4721 |
NFNet F0 nfnet-f0 |
PyTorch |
83.34% |
96.56% |
24.8053 |
71.4444 |
DenseNet 121, alpha=0.125 octave-densenet-121-0.125 |
MXNet |
76.07% |
93.05% |
4.883 |
7.977 |
RegNetX-3.2GF regnetx-3.2gf |
PyTorch |
78.17% |
94.08% |
6.3893 |
15.2653 |
ResNet 26, alpha=0.25 octave-resnet-26-0.25 |
MXNet |
76.08% |
92.58% |
3.768 |
15.99 |
ResNet 50, alpha=0.125 octave-resnet-50-0.125 |
MXNet |
78.19% |
93.86% |
7.221 |
25.551 |
ResNet 101, alpha=0.125 octave-resnet-101-0.125 |
MXNet |
79.18% |
94.42% |
13.387 |
44.543 |
ResNet 200, alpha=0.125 octave-resnet-200-0.125 |
MXNet |
79.99% |
94.87% |
25.407 |
64.667 |
ResNeXt 50, alpha=0.25 octave-resnext-50-0.25 |
MXNet |
78.77% |
94.18% |
6.444 |
25.02 |
ResNeXt 101, alpha=0.25 octave-resnext-101-0.25 |
MXNet |
79.56% |
94.44% |
11.521 |
44.169 |
SE-ResNet 50, alpha=0.125 octave-se-resnet-50-0.125 |
MXNet |
78.71% |
94.09% |
7.246 |
28.082 |
open-closed-eye-0001 open-closed-eye-0001 |
PyTorch |
95.84% |
0.0014 |
0.0113 |
|
RepVGG A0 repvgg-a0 |
PyTorch |
72.40% |
90.49% |
2.7286 |
8.3094 |
RepVGG B1 repvgg-b1 |
PyTorch |
78.37% |
94.09% |
23.6472 |
51.8295 |
RepVGG B3 repvgg-b3 |
PyTorch |
80.50% |
95.25% |
52.4407 |
110.9609 |
ResNeSt 50 resnest-50-pytorch |
PyTorch |
81.11% |
95.36% |
10.8148 |
27.4493 |
ResNet 18 resnet-18-pytorch |
PyTorch |
69.75% |
89.09% |
3.637 |
11.68 |
ResNet 34 resnet-34-pytorch |
PyTorch |
73.30% |
91.42% |
7.3409 |
21.7892 |
ResNet 50 resnet-50-pytorch |
PyTorch |
76.13% |
92.86% |
8.216 |
25.53 |
ResNet 50 resnet-50-caffe2 |
Caffe2 |
76.38% |
93.19% |
8.216 |
25.53 |
ResNet 50 resnet-50-tf |
TensorFlow |
76.17% |
92.98% |
8.2164 |
25.53 |
ReXNet V1 x1.0 rexnet-v1-x1.0 |
PyTorch |
77.86% |
93.87% |
0.8325 |
4.7779 |
SE-Inception se-inception |
Caffe |
76.00% |
92.96% |
4.091 |
11.922 |
SE-ResNet 50 se-resnet-50 |
Caffe |
77.60% |
93.85% |
7.775 |
28.061 |
SE-ResNet 101 se-resnet-101 |
Caffe |
78.25% |
94.21% |
15.239 |
49.274 |
SE-ResNet 152 se-resnet-152 |
Caffe |
78.51% |
94.45% |
22.709 |
66.746 |
SE-ResNeXt 50 se-resnext-50 |
Caffe |
78.97% |
94.63% |
8.533 |
27.526 |
SE-ResNeXt 101 se-resnext-101 |
Caffe |
80.17% |
95.19% |
16.054 |
48.886 |
Shufflenet V2 x1.0 shufflenet-v2-x1.0 |
PyTorch |
69.36% |
88.32% |
0.2957 |
2.2705 |
SqueezeNet v1.0 squeezenet1.0 |
Caffe |
57.68% |
80.38% |
1.737 |
1.248 |
SqueezeNet v1.1 squeezenet1.1 |
Caffe |
58.38% |
81.00% |
0.785 |
1.236 |
SqueezeNet v1.1 squeezenet1.1-caffe2 |
Caffe2 |
56.50% |
79.58% |
0.785 |
1.236 |
VGG 16 vgg16 |
Caffe |
70.97% |
89.88% |
30.974 |
138.36 |
VGG 19 vgg19 |
Caffe |
71.06% |
89.83% |
39.3 |
143.667 |
VGG 19 vgg19-caffe2 |
Caffe2 |
71.06% |
89.83% |
39.3 |
143.667 |
Semantic Segmentation¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
DeepLab V3 deeplabv3 |
TensorFlow |
66.85% |
11.469 |
23.819 |
HRNet V2 C1 Segmentation hrnet-v2-c1-segmentation |
PyTorch |
77.69% |
81.993 |
66.4768 |
Fastseg MobileV3Large LR-ASPP, F=128 fastseg-large |
PyTorch |
72.67% |
140.9611 |
3.2 |
Fastseg MobileV3Small LR-ASPP, F=128 fastseg-small |
PyTorch |
67.15% |
69.2204 |
1.1 |
PSPNet R-50-D8 pspnet-pytorch |
PyTorch |
70.60% |
357.1719 |
46.5827 |
Instance Segmentation¶
Model Name Model Version |
Implementation |
Accuracy Metric 1 |
Accuracy Metric 2 |
GFlops |
mParams |
---|---|---|---|---|---|
Mask R-CNN Inception ResNet V2 mask_rcnn_inception_resnet_v2_atrous_coco |
TensorFlow |
39.86% |
35.36% |
675.314 |
92.368 |
Mask R-CNN Inception V2 mask_rcnn_inception_v2_coco |
TensorFlow |
27.12% |
21.48% |
54.926 |
21.772 |
Mask R-CNN ResNet 50 mask_rcnn_resnet50_atrous_coco |
TensorFlow |
29.75% |
27.46% |
294.738 |
50.222 |
Mask R-CNN ResNet 101 mask_rcnn_resnet101_atrous_coco |
TensorFlow |
34.92% |
31.30% |
674.58 |
69.188 |
YOLACT ResNet 50 FPN yolact-resnet50-fpn-pytorch |
PyTorch |
28.00% |
30.69% |
118.575 |
36.829 |
3D Semantic Segmentation¶
Model Name Model Version |
Implementation |
Mean Accuracy |
Median Accuracy |
GFlops |
mParams |
---|---|---|---|---|---|
Brain Tumor Segmentation brain-tumor-segmentation-0001 |
MXNet |
92.40% |
93.17% |
409.996 |
38.192 |
Brain Tumor Segmentation 2 brain-tumor-segmentation-0002 |
PyTorch |
91.48% |
92.70% |
300.801 |
4.51 |
Object Detection¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
CTPN ctpn |
TensorFlow |
73.67% |
55.813 |
17.237 |
CenterNet (CTDET with DLAV0) 384x384 ctdet_coco_dlav0_384 |
ONNX |
41.61% |
34.994 |
17.911 |
CenterNet (CTDET with DLAV0) 512x512 ctdet_coco_dlav0_512 |
ONNX |
44.28% |
62.211 |
17.911 |
EfficientDet-D0 efficientdet-d0-tf |
TensorFlow |
31.95% |
2.54 |
3.9 |
EfficientDet-D1 efficientdet-d1-tf |
TensorFlow |
37.54% |
6.1 |
6.6 |
FaceBoxes faceboxes-pytorch |
PyTorch |
83.57% |
1.8975 |
1.0059 |
Face Detection Retail face-detection-retail-0044 |
Caffe |
83.00% |
1.067 |
0.588 |
Faster R-CNN with Inception-ResNet v2 faster_rcnn_inception_resnet_v2_atrous_coco |
TensorFlow |
40.69% |
30.687 |
13.307 |
Faster R-CNN with Inception v2 faster_rcnn_inception_v2_coco |
TensorFlow |
26.24% |
30.687 |
13.307 |
Faster R-CNN with ResNet 50 faster_rcnn_resnet50_coco |
TensorFlow |
31.09% |
57.203 |
29.162 |
Faster R-CNN with ResNet 101 faster_rcnn_resnet101_coco |
TensorFlow |
35.72% |
112.052 |
48.128 |
MobileFace Detection V1 mobilefacedet-v1-mxnet |
MXNet |
78.75% |
3.5456 |
7.6828 |
MTCNN mtcnn (mtcnn-p) |
Caffe |
62.26% |
3.366 |
0.007 |
MTCNN mtcnn (mtcnn-r) |
Caffe |
62.26% |
0.003 |
0.1 |
MTCNN mtcnn (mtcnn-o) |
Caffe |
62.26% |
0.026 |
0.389 |
Pelee pelee-coco |
Caffe |
21.98% |
1.290 |
5.98 |
RetinaFace with ResNet 50 retinaface-resnet50-pytorch |
PyTorch |
91.78% |
88.8627 |
27.2646 |
RetinaNet with Resnet 50 retinanet-tf |
TensorFlow |
33.15% |
238.9469 |
64.9706 |
R-FCN with Resnet-101 rfcn-resnet101-coco-tf |
TensorFlow |
45.02% |
53.462 |
171.85 |
SSD 300 ssd300 |
Caffe |
87.09% |
62.815 |
26.285 |
SSD 512 ssd512 |
Caffe |
91.07% |
180.611 |
27.189 |
SSD with MobileNet mobilenet-ssd |
Caffe |
67.00% |
2.316 |
5.783 |
SSD with MobileNet ssd_mobilenet_v1_coco |
TensorFlow |
23.32% |
2.494 |
6.807 |
SSD with MobileNet FPN ssd_mobilenet_v1_fpn_coco |
TensorFlow |
35.55% |
123.309 |
36.188 |
SSD with MobileNet V2 ssd_mobilenet_v2_coco |
TensorFlow |
24.95% |
3.775 |
16.818 |
SSD lite with MobileNet V2 ssdlite_mobilenet_v2 |
TensorFlow |
24.29% |
1.525 |
4.475 |
SSD with ResNet-50 V1 FPN ssd_resnet50_v1_fpn_coco |
TensorFlow |
38.46% |
178.6807 |
59.9326 |
SSD with ResNet 34 1200x1200 ssd-resnet34-1200-onnx |
PyTorch |
39.28% |
433.411 |
20.058 |
Ultra Lightweight Face Detection RFB 320 ultra-lightweight-face-detection-rfb-320 |
PyTorch |
84.78% |
0.2106 |
0.3004 |
Ultra Lightweight Face Detection slim 320 ultra-lightweight-face-detection-slim-320 |
PyTorch |
83.32% |
0.1724 |
0.2844 |
Vehicle License Plate Detection Barrier vehicle-license-plate-detection-barrier-0123 |
TensorFlow |
99.52% |
0.271 |
0.547 |
YOLO v1 Tiny yolo-v1-tiny-tf |
TensorFlow.js |
54.79% |
6.9883 |
15.8587 |
YOLO v2 Tiny yolo-v2-tiny-tf |
Keras |
29.12% |
5.4236 |
11.2295 |
YOLO v2 yolo-v2-tf |
Keras |
56.48% |
63.0301 |
50.9526 |
YOLO v3 yolo-v3-tf |
Keras |
67.72% |
65.9843 |
61.9221 |
YOLO v3 Tiny yolo-v3-tiny-tf |
Keras |
39.70% |
5.582 |
8.848 |
YOLO v4 yolo-v4-tf |
Keras |
77.40% |
129.5567 |
64.33 |
YOLO v4 Tiny yolo-v4-tiny-tf |
Keras |
0.463% |
6.9289 |
6.0535 |
Face Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
FaceNet facenet-20180408-102900 |
TensorFlow |
99.14% |
2.846 |
23.469 |
LResNet100E-IR,ArcFace@ms1m-refine-v2 face-recognition-resnet100-arcface-onnx |
MXNet |
99.68% |
24.2115 |
65.1320 |
SphereFace Sphereface |
Caffe |
98.83% |
3.504 |
22.671 |
Human Pose Estimation¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
human-pose-estimation-3d-0001 human-pose-estimation-3d-0001 |
PyTorch |
100.45mm |
18.998 |
5.074 |
single-human-pose-estimation-0001 single-human-pose-estimation-0001 |
PyTorch |
69.05% |
60.125 |
33.165 |
higher-hrnet-w32-human-pose-estimation higher-hrnet-w32-human-pose-estimation |
PyTorch |
64.64% |
92.8364 |
28.6180 |
Monocular Depth Estimation¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
midasnet midasnet |
PyTorch |
0.07071 |
207.25144 |
104.081 |
FCRN ResNet50-Upproj fcrn-dp-nyu-depth-v2-tf |
TensorFlow |
0.573 |
63.5421 |
34.5255 |
Image Inpainting¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
GMCNN Inpainting gmcnn-places2-tf |
TensorFlow |
33.47Db |
691.1589 |
12.7773 |
Style Transfer¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
fast-neural-style-mosaic-onnx fast-neural-style-mosaic-onnx |
ONNX |
12.04dB |
15.518 |
1.679 |
Action Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
RGB-I3D, pretrained on ImageNet i3d-rgb-tf |
TensorFlow |
86.01% |
278.9815 |
12.6900 |
common-sign-language-0001 common-sign-language-0001 |
PyTorch |
93.58% |
4.2269 |
4.1128 |
Colorization¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
colorization-v2 colorization-v2 |
PyTorch |
26.99dB |
83.6045 |
32.2360 |
colorization-siggraph colorization-siggraph |
PyTorch |
27.73dB |
150.5441 |
34.0511 |
Sound Classification¶
Model Name Model Version |
Implementation |
Top 1 Accuracy |
Top 5 Accuracy |
GFlops |
mParams |
---|---|---|---|---|---|
ACLNet aclnet |
PyTorch |
86% |
92% |
1.4 |
2.7 |
ACLNet-int8 aclnet-int8 |
PyTorch |
87% |
93% |
1.41 |
2.71 |
Speech Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
DeepSpeech V0.6.1 mozilla-deepspeech-0.6.1 |
TensorFlow |
7.55% |
0.0472 |
47.2 |
DeepSpeech V0.8.2 mozilla-deepspeech-0.8.2 |
TensorFlow |
6.13% |
0.0472 |
47.2 |
QuartzNet quartznet-15x5-en |
Pytorch |
3.86% |
2.4195 |
18.8857 |
Image Translation¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
CoCosNet cocosnet |
PyTorch |
12.93dB |
1080.7032 |
167.9141 |
Optical Character Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
license-plate-recognition-barrier-0007 license-plate-recognition-barrier-0007 |
TensorFlow |
98% |
0.347 |
1.435 |
Place Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
NetVLAD netvlad-tf |
TensorFlow |
82.03% |
36.6374 |
149.0021 |
Deblurring¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
DeblurGAN-v2 deblurgan-v2 |
PyTorch |
28.25Db |
80.8919 |
2.1083 |
Salient object detection¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
F3Net f3net |
PyTorch |
84.21% |
31.2883 |
25.2791 |
Text Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
Resnet-FC text-recognition-resnet-fc |
PyTorch |
90.94% |
40.3704 |
177.9668 |
Text to Speech¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
ForwardTacotron forward-tacotron-duration-prediction |
PyTorch |
6.66 |
13.81 |
|
ForwardTacotron forward-tacotron-regression |
PyTorch |
4.91 |
3.05 |
|
WaveRNN wavernn-upsampler |
PyTorch |
0.37 |
0.4 |
|
WaveRNN wavernn-rnn |
PyTorch |
0.06 |
3.83 |
Named Entity Recognition¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
bert-base-NER bert-base-ner |
PyTorch |
94.45% |
22.3874 |
107.4319 |
Vehicle Reidentification¶
Model Name Model Version |
Implementation |
Accuracy |
GFlops |
mParams |
---|---|---|---|---|
vehicle-reid-0001 vehicle-reid-0001 |
PyTorch |
96.31% |
2.643 |
2.183 |
See Also¶
Legal Information¶
Caffe, Caffe2, Keras, MXNet, PyTorch, and TensorFlow are trademarks or brand names of their respective owners. All company, product and service names used in this website are for identification purposes only. Use of these names,trademarks and brands does not imply endorsement.