LEIP Zoo

The LEIP SDK includes a CLI tool to access the Latent AI Model Zoo. The model zoo is an open source repository and contains some common pre-trained model architectures that you are able to access and either run as is, or use in the LEIP flow to further optimize. Each model will also include the source code and configuration, as well as test and train data sets when possible.

CLI Usage

CODE

# You can access the module specific help command
leip zoo --help

# Run the leip zoo list command
leip zoo list

Zoo Models:
            model_id               model_name                           variant_id
   audio-recognition        Audio Recognition                          tf-baseline
           lenet_gtc   LeNet (Training Aware)                       high_precision
           lenet_gtc   LeNet (Training Aware)                        low_precision
         mobilenetv1             Mobilenet V1         keras-open-images-10-classes
         mobilenetv1             Mobilenet V1                       keras-imagenet
         mobilenetv2             Mobilenet V2         keras-open-images-10-classes
         mobilenetv2             Mobilenet V2                       keras-imagenet
         mobilenetv2             Mobilenet V2       pytorch-open-images-10-classes
         mobilenetv2             Mobilenet V2   pytorch-jit-open-images-10-classes
         resnetv2-50              Resnetv2-50         keras-open-images-10-classes
         resnetv2-50              Resnetv2-50                       keras-imagenet
               vgg16                    VGG16         keras-open-images-10-classes
               vgg16                    VGG16                       keras-imagenet
         inceptionv3             Inception V3         keras-open-images-10-classes
         inceptionv3             Inception V3                       keras-imagenet
            xception                 Xception         keras-open-images-10-classes
            xception                 Xception                       keras-imagenet

Zoo Datasets:
               dataset_id             dataset_name     variant_id
   open-images-10-classes   Open Images 10-Classes          train
   open-images-10-classes   Open Images 10-Classes           eval
           pascal-voc2007          Pascal VOC 2007   full-dataset
           pascal-voc2007          Pascal VOC 2007          train
           pascal-voc2007          Pascal VOC 2007           eval
   google-speech-commands   Google Speech Commands          v0.02
   google-speech-commands   Google Speech Commands           eval
   google-speech-commands   Google Speech Commands          train
                    mnist                    MNIST           eval

# Download a model and a variant
leip zoo download --model_id mobilenetv2 --variant_id keras-imagenet

# Note where the model is saved locally...
    Downloading model mobilenetv2 variant keras-imagenet ...
    Archive:  /shared-workdir/workspace/models/mobilenetv2/keras-imagenet/weights.zip
      inflating: model.h5
      inflating: class_names.txt
    Finished downloading model mobilenetv2 variant keras-imagenet ...
    Downloaded files are in workspace/models/mobilenetv2/keras-imagenet

# Download a test dataset
leip zoo download --dataset_id open-images-10-classes --variant_id eval

For a detailed explanation of each option see the CLI Reference for LEIP Zoo.