Note: This repo draws material directly from this Medium article and the IBM MAX model documentation.
These APIs will only be running from March 11th, 2019 to March 22nd, 2019, or until compute credits are depleted.
To use the platform:
- Download and install Postman
- Download ../creative-platform.postman_collection.json and open it in Postman.
- On the left, select the app you want to use, e.g. style-transfer-mosaic
- Go to body
- Next to image, click and select the image you wish to submit to the app
- Click Send
- If you get an error indicating that Postman could not get a response, try resizing the image to be smaller
To run your own implementation of the models on Google Cloud Platform's Kubernetes Engine, follow the steps below.
If you do not have accounts with these services, create them. You will need compute credits for Google Cloud Platform.
If you wish to create an API for use with the platform, you can use this framework to transform your model into an API, though the author notes it may not be ideal for production. The code for the app has been cloned to code/keras-app.
The framework mentioned above was used the create the Docker image of the Image Net Classifier. From here on out, we will assume that any Docker images of interest are already APIs and available on DockerHub or GitHub.
Open your Google Cloud Platform Console and go to Kubernetes Engine.
Create a cluster, selecting 2 nodes and 4vCPUs with 15 GBs of RAM.
Open the cloud shell from the console.
We will be running the following models in our platform:
-
Image Net Classifier As implemented by Gus Cavanaugh and Adrian Rosebrock
To deploy the Image Net classifier, run
kubectl run keras-app --image=oazarate/keras-app --port 5000
Then check the status with kubectl get pods until it returns READY
kubectl get pods
Once it is ready, expose the port to make it publicly available
kubectl expose deployment keras-app --type=LoadBalancer --port 80 --target-port 5000
Then run kubectl get service to find the external IP.
This IP directs to the app and is later used in the creation of the json object for Postman.
To use the IBM MAX models, use:
kubectl apply -f https://raw.githubusercontent.com/IBM/MAX-Fast-Neural-Style-Transfer/master/max-fast-neural-style-transfer.yaml
However, as before, the app is only available locally. In this case, we create a service called my-service that exposes it
kubectl expose deployment max-fast-neural-style-transfer --type=LoadBalancer --name=my-service
You can the find the external IP of the service with kubectl get service