跳至主要内容

Models for Google Coral Edge TPU

In the lists below, each "Edge TPU model" link provides a .tflite file that is pre-compiled to run on the Edge TPU. You can run these models on your Coral device using our example code. (Remember to download the model's corresponding labels file.)

 To get Google Coral Edge TPU products at Gravitylink online store:
 https://store.gravitylink.com/global




Coral Dev board  $149.99
USB Accelerator  $74.99
Mini PCIe Accelerator  $34.99
M.2 Accelerator A+E key  $34.99
M.2 Accelerator B+M key  $34.99
System-on-Module (SoM)
$114.99
  
For many of the models, we've also provided a link for "All model files," which is an archive file that includes the following:
  • Trained model checkpoints
  • Frozen graph for the trained model
  • Eval graph text protos (to be easily viewed)
  • Info file containing input and output information
  • Quantized TensorFlow Lite model that runs on CPU (included with classification models only)

Download this "All model files" archive to get the checkpoint file you'll need if you want to use the model as your basis for transfer-learning, as shown in the tutorials to retrain a classification model and retrain an object detection model.
If you'd like to download all models at once, you can clone our Git repo https://github.com/google-coral/edgetpu and then find the models in test_data/.
Notice: These are not production-quality models; they are for demonstration purposes only.
To build your own model for the Edge TPU, you must use the Edge TPU Compiler.
All models trained on ImageNet used the ILSVRC2012 dataset.

Image classification


EfficientNet-EdgeTpu (S)
Recognizes 1,000 types of objects
Dataset: ImageNet
Input size: 224x224


EfficientNet-EdgeTpu (M)
Recognizes 1,000 types of objects
Dataset: ImageNet
Input size: 240x240

EfficientNet-EdgeTpu (L)
Recognizes 1,000 types of objects
Dataset: ImageNet
Input size: 300x300

MobileNet V1 (ImageNet)

    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 224x224

    MobileNet V2 (ImageNet)
    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 224x224

    MobileNet V2 (iNat insects)
    Recognizes 1,000+ types of insect
    Dataset: iNaturalist
    Input size: 224x224

    MobileNet V2 (iNat plants)
    Recognizes 2,000+ types of plants
    Dataset: iNaturalist
    Input size: 224x224

    MobileNet V2 (iNat birds)
    Recognizes 900+ types of birds
    Dataset: iNaturalist
    Input size: 224x224

    Inception V1 (ImageNet)
    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 224x224

    Inception V2 (ImageNet)
    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 224x224

    Inception V3 (ImageNet)
    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 299x299

    Inception V4 (ImageNet)
    Recognizes 1,000 types of objects
    Dataset: ImageNet
    Input size: 299x299

    Object detection


    MobileNet SSD v1 (COCO)
    Detects the location of 90 types objects
    Dataset: COCO
    Input size: 300x300
    MobileNet SSD v2 (COCO)

    Detects the location of 90 types objects
    Dataset: COCO
    Input size: 300x300
    MobileNet SSD v2 (Faces)

    Detects the location of human faces
    Dataset: Open Images v4
    Input size: 320x320
    (Does not require a labels file)

    Semantic segmentation


    MobileNet v2 DeepLab v3 (0.5 depth multiplier)
    Recognizes and segments 20 types of objects
    Dataset: PASCAL VOC 2012
    Input size: 513x513
    Depth multiplier: 0.5

    MobileNet v2 DeepLab v3 (1.0 depth multiplier)
    Recognizes and segments 20 types of objects
    Dataset: PASCAL VOC 2012
    Input size: 513x513
    Depth multiplier: 1.0

    On-device retraining (classification)


    MobileNet v1 embedding extractor
    This model is compiled with the last fully-connected layer removed so that it can be used as an embedding extractor for on-device transfer-learning with the SoftmaxRegression API. This model does not perform classifications on its own, and must be paired with the SoftmaxRegression API.

    MobileNet v1 with L2-norm
    This is a modified version of MobileNet v1 that includes an L2-normalization layer and other changes to be compatible with the ImprintingEngine API. It's built for the Edge TPU but the last fully-connected layer executes on the CPU to enable retraining.



    评论

    发表评论

    此博客中的热门博文

    10 questions you should know about Google Edge TPU

    1. What is the Edge TPU? The Edge TPU is a small ASIC designed by Google that provides high performance ML inferencing for low-power devices. For example, it can execute state-of-the-art mobile vision models such as MobileNet V2 at almost 400 FPS, in a power efficient manner. Google offers multiple products that include the Edge TPU built-in. Two Edge TPU chips on the head of a US penny 2. What machine learning frameworks does the Edge TPU support? TensorFlow Lite only. 3. What type of neural networks does the Edge TPU support? The first-generation Edge TPU is capable of executing deep feed-forward neural networks (DFF) such as convolutional neural networks (CNN), making it ideal for a variety of vision-based ML applications. 4. How do I create a TensorFlow Lite model for the Edge TPU? You need to convert your model to TensorFlow Lite and it must be quantized using either quantization-aware training (recommended) or full integer post-training quantization. (To create a compatible model...
    发表评论
    阅读全文

    How to Retrain an object detection model

    This tutorial shows you how to retrain an object detection model to recognize a new set of classes. You'll use a technique called transfer learning to retrain an existing model and then compile it to run on an Edge TPU device—you can use the retrained model with either the Coral Dev Board or the Coral USB Accelerator. Specifically, this tutorial shows you how to retrain a MobileNet V1 SSD model (originally trained to detect 90 objects from the COCO dataset) so that it detects two pets: Abyssinian cats and American Bulldogs (from the Oxford-IIIT Pets Dataset). But you can reuse these procedures with your own image dataset, and with a different pre-trained model. The steps below show you how to perform transfer-learning using either last-layers-only or full-model retraining. Most of the steps are the same; just keep an eye out for the different commands depending on the technique you desire. Note: These instructions do not require deep experience with TensorFlow o...
    发表评论
    阅读全文

    Introducing Google Coral Dev Board

    Introducing Google Coral Dev Board (Part 1) Introducing Google Coral Dev Board (Part 2) Introducing Google Coral Dev Board (Part 3) Introducing Google Coral Dev Board (Part 4)
    发表评论
    阅读全文