MODELS_cn.md

English | 简体中文

本文列出了所有经过验证的 OpenVINO™ integration with TensorFlow 支持的模型。随着我们支持越来越多的算子和模型，该列表将会持续更新。

TensorFlow-Slim 图像分类库

模型名称	支持的设备
Inception V3	CPU, iGPU, MYRIAD, VAD-M
Inception_V4	CPU, iGPU, MYRIAD, VAD-M
Resnet V1 50	CPU, iGPU, MYRIAD, VAD-M
Resnet V2 152	CPU, iGPU, MYRIAD, VAD-M
Resnet V2 50	CPU, iGPU, MYRIAD, VAD-M
VGG 16	CPU, iGPU, MYRIAD, VAD-M
VGG 19	CPU, iGPU, MYRIAD, VAD-M
MobileNet_v1_1.0_224	CPU, iGPU, MYRIAD, VAD-M
MobileNet_v2_1.4_224	CPU, iGPU, MYRIAD, VAD-M
CifarNet	CPU, iGPU, MYRIAD, VAD-M
LeNet	CPU, iGPU, MYRIAD, VAD-M

TensorFlow-Slim 模型链接仅包含预训练checkpoint文件。您可以参考 TensorFlow-Slim 模型说明页面，以便运行推理或冻结模型。（不提供针对 CifarNet 和 LeNet 的预训练checkpoint文件。）

TensorFlow 对象检测 Model Zoo

模型名称	支持的设备
faster_rcnn_inception_resnet_v2_atrous_coco	CPU, iGPU, MYRIAD, VAD-M
faster_rcnn_inception_v2_coco	CPU, iGPU, MYRIAD, VAD-M
faster_rcnn_resnet50_coco	CPU, iGPU, MYRIAD, VAD-M
faster_rcnn_resnet101_coco	CPU, iGPU, MYRIAD, VAD-M
faster_rcnn_resnet50_lowproposals_coco	CPU, iGPU, MYRIAD, VAD-M
ssd_inception_v2	CPU, iGPU, MYRIAD, VAD-M
ssd_mobilenet_v1	CPU, iGPU, MYRIAD, VAD-M
ssd_mobilenet_v1_fpn	CPU, iGPU, MYRIAD, VAD-M
ssd_mobilenet_v2	CPU, iGPU, MYRIAD, VAD-M
ssd_resnet_50_fpn	CPU, iGPU, MYRIAD, VAD-M
ssdlite_mobilenet_v2	CPU, iGPU, MYRIAD, VAD-M
mask_rcnn_inception_resnet_v2_atrous_coco	CPU, iGPU, MYRIAD, VAD-M
mask_rcnn_inception_v2_coco	CPU, iGPU, MYRIAD, VAD-M

为这些模型提供了预训练frozen模型。

TensorFlow Keras 应用

模型名称	支持的设备
DenseNet121	CPU, iGPU, MYRIAD, VAD-M
DenseNet169	CPU, iGPU, MYRIAD, VAD-M
DenseNet201	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB0	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB1	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB2	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB3	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB4	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB5	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB6	CPU, iGPU, MYRIAD, VAD-M
EfficientnetB7	CPU, iGPU, MYRIAD, VAD-M
InceptionV3	CPU, iGPU, MYRIAD, VAD-M
NASNetLarge	CPU, iGPU, MYRIAD, VAD-M
NASNetMobile	CPU, iGPU, MYRIAD, VAD-M
ResNet50v2	CPU, iGPU, MYRIAD, VAD-M

请参考 Keras 应用页面上的说明，了解更多关于如何使用这些预训练模型的信息。

TensorFlow-Hub 模型

模型名称	支持的设备
albert_en_base	CPU, iGPU, MYRIAD
albert_en_preprocess	CPU, iGPU, MYRIAD
albert_en_xxlarge	CPU, iGPU, MYRIAD
bert_en_cased_L-12_H-768_A-12	CPU, iGPU, MYRIAD
bert_en_cased_L-24_H-1024_A-16	CPU, iGPU, MYRIAD
bert_en_uncased_L-12_H-768_A-12	CPU, iGPU, MYRIAD
bert_en_uncased_L-24_H-1024_A-16	CPU, iGPU, MYRIAD
bert_en_uncased_preprocess	CPU, iGPU, MYRIAD
bert_en_wwm_uncased_L-24_H-1024_A-16	CPU, iGPU, MYRIAD
bert_multi_cased_preprocess	CPU, iGPU, MYRIAD
bert_zh_L-12_H-768_A-12	CPU, iGPU, MYRIAD
bert_zh_preprocess	CPU, iGPU, MYRIAD
bit​/m-r101x1	CPU, iGPU, MYRIAD
bit​/m-r101x3	CPU, iGPU, MYRIAD
bit​/m-r50x1	CPU, iGPU, MYRIAD
bit​/m-r50x1​/ilsvrc2012_classification	CPU, iGPU, MYRIAD
bit​/m-r50x3	CPU, iGPU, MYRIAD
bit​/s-r50x1	CPU, iGPU, MYRIAD
centernet​/hourglass_512x512	iGPU, MYRIAD
centernet​/hourglass_512x512_kpts	CPU, iGPU, MYRIAD
centernet​/resnet50v1_fpn_512x512	iGPU, MYRIAD
cropnet​/classifier​/cassava_disease_V1	CPU, iGPU, MYRIAD
efficientdet​/d0	CPU, iGPU, MYRIAD
efficientdet​/d4	CPU, iGPU, MYRIAD
efficientdet​/d7	CPU, iGPU, MYRIAD
efficientdet​/lite0​/detection	CPU, iGPU, MYRIAD
efficientdet​/lite0​/feature-vector	CPU, iGPU, MYRIAD
efficientnet​/b0​/classification	CPU, iGPU, MYRIAD
efficientnet​/b0​/feature-vector	CPU, iGPU, MYRIAD
efficientnet​/b3​/feature-vector	CPU, iGPU, MYRIAD
efficientnet​/b4​/feature-vector	CPU, iGPU, MYRIAD
efficientnet​/b7​/classification	CPU, iGPU, MYRIAD
efficientnet​/b7​/feature-vector	CPU, iGPU, MYRIAD
electra_small	CPU, iGPU, MYRIAD
esrgan-tf2	CPU, iGPU, MYRIAD
experts​/bert​/wiki_books	CPU, iGPU, MYRIAD
faster_rcnn​/inception_resnet_v2_1024x1024	CPU, iGPU, MYRIAD
faster_rcnn​/inception_resnet_v2_640x640	CPU, iGPU, MYRIAD
faster_rcnn​/resnet50_v1_640x640	CPU, iGPU, MYRIAD
imagenet​/inception_resnet_v2​/classification	CPU, iGPU, MYRIAD
imagenet​/inception_resnet_v2​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/inception_v1​/classification	CPU, iGPU, MYRIAD
imagenet​/inception_v1​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/inception_v2​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/inception_v3​/classification	CPU, iGPU, MYRIAD
imagenet​/inception_v3​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v1_025_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v1_050_160​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v1_100_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v1_100_224​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_035_128​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_035_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_035_96​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_050_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_075_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_100_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_100_224​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_130_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_140_224​/classification	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v2_140_224​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/mobilenet_v3_small_100_224​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/pnasnet_large​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/resnet_v1_50​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/resnet_v2_152​/feature_vector	CPU, iGPU, MYRIAD
imagenet​/resnet_v2_50​/classification	CPU, iGPU, MYRIAD
imagenet​/resnet_v2_50​/feature_vector	CPU, iGPU, MYRIAD
inaturalist​/inception_v3​/feature_vector	CPU, iGPU, MYRIAD
LaBSE	CPU, iGPU, MYRIAD
mask_rcnn​/inception_resnet_v2_1024x1024	CPU, iGPU, MYRIAD
movenet​/singlepose​/lightning	CPU, iGPU, MYRIAD
MuRIL	CPU, iGPU, MYRIAD
nnlm-de-dim50	CPU, iGPU, MYRIAD
nnlm-de-dim50-with-normalization	CPU, iGPU, MYRIAD
nnlm-en-dim128	CPU, iGPU, MYRIAD
nnlm-en-dim128-with-normalization	CPU, iGPU, MYRIAD
nnlm-en-dim50	CPU, iGPU, MYRIAD
nnlm-en-dim50-with-normalization	CPU, iGPU, MYRIAD
nnlm-es-dim128	CPU, iGPU, MYRIAD
nnlm-ja-dim128	CPU, iGPU, MYRIAD
nnlm-ja-dim128-with-normalization	CPU, iGPU, MYRIAD
nnlm-ja-dim50	CPU, iGPU, MYRIAD
nonsemantic-speech-benchmark​/trill	CPU, iGPU, MYRIAD
nonsemantic-speech-benchmark​/trill-distilled	CPU, iGPU, MYRIAD
resnet_50​/classification	CPU, iGPU, MYRIAD
resnet_50​/feature_vector	CPU, iGPU, MYRIAD
small_bert​/bert_en_uncased_L-2_H-128_A-2	CPU, iGPU, MYRIAD
small_bert​/bert_en_uncased_L-4_H-512_A-8	CPU, iGPU, MYRIAD
ssd_mobilenet_v2	CPU, iGPU, MYRIAD
ssd_mobilenet_v2​/fpnlite_320x320	CPU, iGPU, MYRIAD
ssd_mobilenet_v2​/fpnlite_640x640	CPU, iGPU, MYRIAD
tf2-preview​/gnews-swivel-20dim	CPU, iGPU, MYRIAD
tf2-preview​/gnews-swivel-20dim-with-oov	CPU, iGPU, MYRIAD
tf2-preview​/inception_v3​/classification	CPU, iGPU, MYRIAD
tf2-preview​/inception_v3​/feature_vector	CPU, iGPU, MYRIAD
tf2-preview​/mobilenet_v2​/classification	CPU, iGPU, MYRIAD
tf2-preview​/mobilenet_v2​/feature_vector	CPU, iGPU, MYRIAD
tf2-preview​/nnlm-en-dim128	CPU, iGPU, MYRIAD
tf2-preview​/nnlm-en-dim128-with-normalization	CPU, iGPU, MYRIAD
tf2-preview​/nnlm-en-dim50	CPU, iGPU, MYRIAD
tf2-preview​/nnlm-es-dim50-with-normalization	CPU, iGPU, MYRIAD
universal-sentence-encoder	CPU, iGPU, MYRIAD
universal-sentence-encoder-cmlm​/multilingual-preprocess	CPU, iGPU, MYRIAD
universal-sentence-encoder-large	CPU, iGPU, MYRIAD
universal-sentence-encoder-multilingual	CPU, iGPU, MYRIAD
universal-sentence-encoder-multilingual-large	CPU, iGPU, MYRIAD
universal-sentence-encoder-multilingual-qa	CPU, iGPU, MYRIAD
universal-sentence-encoder-qa	CPU, iGPU, MYRIAD
vggish	CPU, iGPU, MYRIAD
Wiki-words-250	CPU, iGPU, MYRIAD
Wiki-words-250-with-normalization	CPU, iGPU, MYRIAD
Wiki-words-500-with-normalization	CPU, iGPU, MYRIAD
yamnet	CPU, iGPU, MYRIAD

其他模型

模型名称	支持的设备
3d-pose-baseline	CPU, iGPU, MYRIAD, VAD-M
ACGAN	CPU
adv_inception_v3	CPU
ALBERT	CPU
arttrack-coco-multi	CPU, iGPU, MYRIAD
arttrack-mpii-single	CPU, iGPU, MYRIAD
BERT_LARGE	CPU
context_rcnn_resnet101_snapshot_serenget	CPU
cpm-person	CPU, iGPU, MYRIAD, VAD-M
cpm-pose	CPU, iGPU, VAD-M
CRNN	CPU, iGPU, MYRIAD, VAD-M
ctpn	CPU, iGPU, MYRIAD, VAD-M
deeplab	CPU
deeplabv3	CPU, MYRIAD, VAD-M
densenet161	CPU, iGPU, MYRIAD, VAD-M
dilation	CPU, iGPU
east_resnet_v1_50	CPU, iGPU, MYRIAD, VAD-M
efficientdet-d0_frozen	CPU,  iGPU,  MYRIAD
EfficientDet-D0-512x512	CPU
efficientdet-d1_frozen	CPU,  iGPU,  MYRIAD
EfficientDet-D1-640x640	CPU
EfficientDet-D2-768x768	CPU
EfficientDet-D3-896x896	CPU
EfficientDet-D4-1024x1024	CPU
EfficientDet-D5-1280x1280	CPU
EfficientDet-D6-1280x1280	CPU
EfficientDet-D7-1536x1536	CPU
efficientnet-b0_auto_aug	CPU, iGPU, MYRIAD, VAD-M
efficientnet-b7_auto_aug	CPU, iGPU, MYRIAD, VAD-M
ens3_adv_inception_v3	CPU
facenet-20180408-102900	CPU, iGPU, MYRIAD
faster_rcnn_inception_resnet_v2_atrous_lowproposals_coco	CPU
faster_rcnn_inception_resnet_v2_atrous_oid	CPU
faster_rcnn_nas_coco	CPU
faster_rcnn_nas_coco_2018_01_28	CPU, VAD-M
faster_rcnn_nas_lowproposals_coco	CPU, MYRIAD VAD-M
faster_rcnn_resnet101_ava_v2.1	CPU
faster_rcnn_resnet101_fgvc	CPU
faster_rcnn_resnet101_kitti	CPU
faster_rcnn_resnet101_lowproposals_coco	CPU
faster_rcnn_resnet101_snapshot_serengeti	CPU
faster_rcnn_resnet50_fgvc	CPU
fc_densenet_103	CPU, iGPU, MYRIAD, VAD-M
fcrn-dp-nyu-depth-v2-tf	CPU, iGPU, MYRIAD, VAD-M
googlenet-v1	CPU, iGPU, MYRIAD, VAD-M
googlenet-v2	CPU, iGPU, MYRIAD, VAD-M
googlenet-v3	CPU, iGPU, MYRIAD, VAD-M
googlenet-v4	CPU, iGPU, MYRIAD, VAD-M
handwritten-score-recognition-0003	CPU
i3d-flow	CPU, iGPU, MYRIAD, VAD-M
i3d-rgb	CPU, iGPU, MYRIAD, VAD-M
icnet-camvid-ava-0001	CPU
icnet-camvid-ava-sparse-30-0001	CPU
icnet-camvid-ava-sparse-60-0001	CPU
image-retrieval-0001	CPU
inception_resnet_v2	CPU, iGPU, MYRIAD, VAD-M
license-plate-recognition-barrier-0007	CPU, iGPU, MYRIAD, VAD-M
lm_1b	CPU, iGPU, MYRIAD
mask_rcnn_resnet101_atrous_coco	CPU, iGPU, MYRIAD, VAD-M
mask_rcnn_resnet50_atrous	CPU, iGPU, MYRIAD
mask_rcnn_resnet50_atrous_coco	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v1_0.25_128	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v1_0.50_160	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v1_0.50_224	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v1_1.0_224	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v2_1.0_224	CPU, iGPU, MYRIAD, VAD-M
mobilenet_v2_fpn_ssdlite_crossroad	CPU, iGPU, MYRIAD, VAD-M
mobilenet-v3-large-1.0-224-tf	CPU, iGPU, MYRIAD, VAD-M
mobilenet-v3-small-1.0-224-tf	CPU, iGPU, MYRIAD, VAD-M
mozilla-deepspeech-0.6.1	CPU,  iGPU, MYRIAD, VAD-M
mozilla-deepspeech-0.7.1	CPU, iGPU, MYRIAD, VAD-M
mozilla-deepspeech-0.8.2	CPU, MYRIAD, VAD-M
NCF-1B	CPU
netvlad	CPU, iGPU, MYRIAD, VAD-M
NiftyNet	CPU, iGPU, MYRIAD, VAD-M
openpose-pose	CPU, iGPU, MYRIAD, VAD-M
person-vehicle-bike-detection-crossroad-yolov3-1020	CPU
pose-ae-multiperson	CPU, iGPU, MYRIAD, VAD-M
pose-ae-refinement	CPU, iGPU, MYRIAD, VAD-M
PRNet	CPU, iGPU, MYRIAD, VAD-M
resnet_50	CPU, iGPU, MYRIAD, VAD-M
resnet_50_v1.5	CPU, iGPU, MYRIAD, VAD-M
resnet_v2_101	CPU, iGPU, MYRIAD, VAD-M
resnet-101	CPU, iGPU, MYRIAD, VAD-M
resnet-152	CPU, iGPU, MYRIAD, VAD-M
ResNet-50_v1.5	CPU
resnet-50-tf	CPU, iGPU, MYRIAD, VAD-M
ResNeXt_101	CPU
ResNext_50	CPU
resnext50v2	CPU,  iGPU,  MYRIAD, VAD-M
R-FCN	CPU
rfcn_resnet101_coco	CPU, iGPU, MYRIAD, VAD-M
se-resnext-50	CPU, iGPU, MYRIAD, VAD-M
SqueezeNet	CPU
squeezenet1.1	CPU, iGPU, MYRIAD, VAD-M
srgan	CPU, iGPU, MYRIAD, VAD-M
ssd_inception_v2_coco	CPU
ssd_mobilenet_v1_0.75_depth_300x300_coco	CPU
ssd_mobilenet_v1_ppn_shared_box_predictor_300x300_coco	CPU
ssd_resnet_101_fpn_oidv4	CPU
ssd_resnet34_1200x1200	CPU
ssd_resnet34_300x300	CPU
ssd_resnet34_fp32_1200x1200_pretrained_model	CPU
SSD_ResNet50_V1_FPN_640x640_RetinaNet50	CPU
ssd_resnet50_v1_fpn_coco	CPU
STN	CPU, iGPU, MYRIAD, VAD-M
text-recognition-0012	CPU
Transformer-LT	CPU
vehicle-attributes-barrier-0103	CPU, iGPU, MYRIAD, VAD-M
vehicle-license-plate-detection-barrier-0106	CPU
vehicle-license-plate-detection-barrier-0123	CPU, iGPU, MYRIAD, VAD-M
vggvox	CPU, iGPU, MYRIAD, VAD-M
wavenet	CPU
wide_deep	CPU
yolo_v2	CPU, iGPU, MYRIAD, VAD-M
yolo_v2_tiny	CPU, iGPU, MYRIAD, VAD-M
yolo_v3_darknet	CPU, iGPU, MYRIAD, VAD-M
yolo-v1-tiny-tf	CPU, iGPU, MYRIAD
yolo-v3	CPU, iGPU, MYRIAD, VAD-M
yolo-v3-tiny-tf	CPU, iGPU, MYRIAD, VAD-M
yolo-v4	CPU, iGPU, MYRIAD, VAD-M

仅针对其中一些模型提供了预训练的冻结模型文件。 其余的，请参考提供的链接。

量化支持 (Preview)

OpenVINO™ integration with TensorFlow 现在支持使用量化感知训练 (QAT) 工具量化的 INT8 模型，例如 OpenVINO™ 神经网络压缩框架 (NNCF) 和 TensorFlow 模型优化工具包 (TFMOT)。 此支持目前处于预览状态，性能优化正在进行中。

可以在 此处 中找到一些使用 NNCF 生成量化模型的示例。

通过在运行推理之前将环境变量“OPENVINO_TF_CONSTANT_FOLDING”设置为 1，一些量化模型可以获得更优的性能。

[注意：NNCF 和 OpenVINO™ integration with TensorFlow 支持的最新 TensorFlow 版本可能不同。 建议用户创建一个单独的虚拟环境来使用 NNCF 量化模型，以避免任何 TensorFlow 版本不兼容问题。 然后可以在兼容 OpenVINO™ integration with TensorFlow 的环境中运行量化模型。 NNCF兼容TensorFlow 2.4.2， OpenVINO™ integration with TensorFlow 兼容 TensorFlow 2.5.1，已经过验证。]