Neuralet > Models > Edge TPU
Edge TPU

Edge TPU

37

Image Classification

  • Model
  • Task
  • Inference Time (ms)
  • Throughput (FPS)
  • Description
  • File
  • Repo or Website
  • Neuralet-OFMClassifier

    Official classifier by Neuralet pre-trained on the Extended-Synthetic-Blurred dataset that classifies masked faces from unmasked ones.

    21

  • InceptionV4

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 299x299 that recognizes 1000 classes.

    10

  • InceptionV3

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 299x299 that recognizes 1000 classes.

    11

  • InceptionV2

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 224x224 that recognizes 1000 classes.

    9

  • InceptionV1

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 224x224 that recognizes 1000 classes.

    11

  • MobileNetV2_iNatBirds

    Official classifier by Coral pre-trained on the iNaturalist dataset (iNat birds) on an input size of 224x224 that recognizes more than 900 types of birds.

    10

  • MobileNetV2_iNatPlants

    Official classifier by Coral pre-trained on the iNaturalist dataset (iNat plants) on an input size of 224x224 that recognizes more than 2000 types of plants.

    9

  • EfficientNet_EdgeTpu_L

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 300x300 that recognizes 1000 classes.

    8

  • EfficientNet_EdgeTpu_M

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 240x240 that recognizes 1000 classes.

    9

  • EfficientNet_EdgeTpu_S

    Official classifier by Coral pre-trained on the ImageNet dataset on an input size of 224x224 that recognizes 1000 classes.

    11

Object Detection

  • Model
  • Task
  • Inference Time (ms)
  • Throughput (FPS)
  • Description
  • File
  • Repo or Website
  • SSD_MobileNetV2_OpenImagesV4

    Official object detector model by Coral pre-trained on the Open Images V4 dataset on an input size of 320x320 that recognizes human face.

    14

  • SSD_MobileNetV2_COCO

    Official object detector model by Coral pre-trained on the MSCOCO dataset on an input size of 300x300 that recognizes 90 classes.

    12

  • SSD_MobileNetV1_COCO

    Official object detector model by Coral pre-trained on the MSCOCO dataset on an input size of 300x300 that recognizes 90 classes.

    12

Pose Estimation

  • Model
  • Task
  • Inference Time (ms)
  • Throughput (FPS)
  • Description
  • File
  • Repo or Website
  • PoseNet_MobileNetV1_1281x721

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 1281x721 and a multiplier of 0.75.

    8

  • PoseNet_MobileNetV1_641x481

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 641x481 and a multiplier of 0.75.

    7

  • PoseNet_MobileNetV1_481x353

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 481x353 and a multiplier of 0.75.

    9

  • PoseNet_ResNet50_960x736

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 960x736 and an output stride of 32.

    8

  • PoseNet_ResNet50_928x672

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 928x672 and an output stride of 16.

    10

  • PoseNet_ResNet50_864x624

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 864x624 and an output stride of 32.

    8

  • PoseNet_ResNet50_768x496

    Official pose estimation model by Coral pre-trained on the MSCOCO dataset on an input resolution of 768x496 and an output stride of 32.

    8

  • PoseNet_ResNet50_640x480

    Official pose estimation model by Google pre-trained on the MSCOCO dataset on an input resolution of 640x480 and an output stride of 16.

    8

  • PoseNet_ResNet50_416x288

    Official pose estimation model by Google pre-trained on the MSCOCO dataset on an input resolution of 416x288 and an output stride of 16.

    8

Semantic Segmentation

  • Model
  • Task
  • Inference Time (ms)
  • Throughput (FPS)
  • Description
  • File
  • Repo or Website
  • UNet_MobileNetV2_256x256

    Official model by Coral that recognizes and segments pets using 3 classes: pixels belonging to a pet, pixels bordering a pet, and background pixels (it does not classify the type of pet), and is pre-trained on the Oxford-IIIT Pet dataset on an input size of 256x256.

    9

  • UNet_MobileNetV2_128x128

    Official model by Coral that recognizes and segments pets using 3 classes: pixels belonging to a pet, pixels bordering a pet, and background pixels (it does not classify the type of pet), and is pre-trained on the Oxford-IIIT Pet dataset on an input size of 128x128.

    9

  • DeepLabV3_MobileNetV2_513x513_1.0

    Official model by Coral that recognizes and segments 20 types of objects and is pre-trained on the PASCAL VOC 2012 dataset on an input size of 513x513 and depth multiplier of 1.0.

    9

  • DeepLabV3_MobileNetV2_513x513_0.5

    Official model by Coral that recognizes and segments 20 types of objects and is pre-trained on the PASCAL VOC 2012 dataset on an input size of 513x513 and depth multiplier of 0.5.

    12

  • BodyPix_ResNet50_960x736

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 960x736 and output stride of 32.

    8

  • BodyPix_ResNet50_928x672

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 928x672 and output stride of 16.

    7

  • BodyPix_ResNet50_864x624

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 864x624 and output stride of 32.

    8

  • BodyPix_ResNet50_768x496

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 768x496 and output stride of 32.

    7

  • BodyPix_ResNet50_640x480

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 640x480, depth multiplier of 0.75, and output stride of 16.

    8

  • BodyPix_ResNet50_416x288

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 416x288, depth multiplier of 0.75, and output stride of 16.

    8

  • BodyPix_MobileNetV1_768x576

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 768x576, depth multiplier of 0.75, and output stride of 16.

    7

  • BodyPix_MobileNetV1_640x480

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 640x480, depth multiplier of 0.75, and output stride of 16.

    8

  • BodyPix_MobileNetV1_480x352

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 480x352, depth multiplier of 0.75, and output stride of 16.

    8

  • BodyPix_MobileNetV1_1280x720

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 1280x720, depth multiplier of 0.75, and output stride of 16.

    6

  • BodyPix_MobileNetV1_1024x768

    Official model by Coral that allows for person and body-part segmentation and is pre-trained with an input size of 1024x768, depth multiplier of 0.75, and output stride of 16.

    11

Get In Touch

Have a question? Send us a message and we will respond as soon as possible.