Deploy TF-Lite model to AWS cloud: ECR, Lambda and API Gateway

This is a classifier for dinosaur and dragon developed with AWS services.

Pretrained Model dino_dragon.tflite

def make_model(input_size=150, learning_rate=0.002, size_inner=64):

    inputs = keras.Input(shape=(input_size, input_size, 3))

    conv1 = tf.keras.layers.Conv2D(filters=32, kernel_size=(3, 3), activation="relu")(inputs)
    pool = keras.layers.MaxPooling2D(pool_size=(2, 2))(conv1)
    flat = keras.layers.Flatten()(pool)
    FC = keras.layers.Dense(size_inner, activation='relu')(flat)
    outputs = keras.layers.Dense(1, activation='sigmoid')(FC)

    model = keras.Model(inputs, outputs)
    
    optimizer = tf.keras.optimizers.SGD(learning_rate=learning_rate, momentum=0.8)
    loss = tf.keras.losses.BinaryCrossentropy()

    model.compile(optimizer=optimizer, loss=loss, metrics=['accuracy'])

    return model

Testing with testing.py

• Ensure the host is correct
• Change the img_url to desired image in the script

python test.py

To Run this APP on your own

1. Train a Tensorflow model and clone this repo
2. Convert your TF model to TF-Lite with convert.py
3. Define the lambda function
   • See lambda_function.py
   • Test with terminal

   python
   >>> import lambda_function
   >>> event = {"url": "{some image url}"}
   >>> lambda_function.lambda_handler(event, None)
   

4. Build the image
   • See Dockerfile
   • Run and test

   docker build -t {image_name} .
   docker run --rm -p 8080:8080 {image_name}
   python test.py
   

5. Publishing to AWS
   • Publish image to ECR

   aws ecr create-repository --repository-name {repo_name}
   aws ecr get-login-password --region {region} | docker login\
       --username AWS \
       --password-stdin {aws_account_id}.dkr.ecr.{region}.amazonaws.com
   docker tag {docker image} {repositoryUri}:{Tag}
   docker push {repositoryUri}:{Tag}
   

   • Create a lambda function using the ECR image
     • Change configuration (needed for model initialization)
       • timeout -> 30s
       • memory -> 1024ME
   • Expose lambda function to API Gateway
     • Create resource
     • Create method
     • Deploy API