Source: Deep Learning on Medium
How to Build an Image Classifier Using Watson AutoAI
Build an image classification model without coding
IBM Watson Studio — AutoAI
AutoAI is a platform that helps you accelerate your AI pipeline through:
- Understanding your data in a better way
- Cleaning the data
- Data preparation
- Feature engineering
- Fine Tuning
- Providing inference APIs
What used to be done in days, now can be done in minutes with great results and without writing a single line of code!
Image classification is one of the tasks in Computer Vision where the goal is to classify an image based on its contents. For example, given the bellow image, the expected output is dog (if the training dataset is for dogs and cats for example).
Let us build the classifier
We will use Stanford Dogs Dataset that contains images of 120 breeds of dogs from around the world.
First, go to Watson AutoAI and create a project.
Click on the project name and you should be directed to the project home page. Click on “Add to project”
Then choose “Visual Recognition Model”
If it is your first time, you need to create a service. Click “here”
Choose a plan that fits your needs.
You have the option of changing the region, plan, resource group, etc. I prefer the default settings.
Click on “Classify Images” to create a classification model
This is the main page for building the model. Click “Browse” on the right side to upload your dataset (zip files where the name of the folder is the name of the class).
I have uploaded three datasets (Dingo, Great Dane, and Siberian Husky) with 50 images in each.
Click “Train” to start training the model.
Once the training is done, you will get a notification. Click “here” to be redirected to the project overview.
A list of information will be provided about the model.
Click on “Test” to inference the model. Upload an image and the model will predict the class for the image. You can choose the displayed classes from the left side along with their confidence.
Finally, click on “Implementation” and you will be provided with APIs to inference the model remotely.