This recording shows the end-to-end demo of the steps below. Feel free to follow if you're stuck. It's unnarrated however.
Head on over to https://cloud.ibm.com/registration to create an IBM Cloud Account. This will allow you to try out all the services and APIs available on the IBM Cloud.
Once you've entered your email you'll be prompted to confirm your email address. Head on over and confirm it.
Login once you're redirected to the IBM Cloud account, you'll be prompted to proceed. Hit Proceed. From here you'll end up at your IBM Dashboard.
The next bit of setup involves creating a Watson Studio service. To do this hit Catalog at the top of the screen.
Hit AI from the menu on the right hand side. Then select Watson Studio.
Then hit Create from the right hand side of the screen.
Awesome! You've now created a Watson Studio environment. Next hit get started to create a Watson Studio Project.
First thing’s first, the easiest way to work with the Object Detection API is to create a new project inside of Watson Studio. Start out by selecting Create a project.
Then select Create an empty project.
IMPORTANT Adding Cloud Object Storage: you need to add this so you can upload images.
Select Add, you'll be taken away from your screen so that you can attach some storage.
Choose the Lite Plan then hit Create.
Then come back into your project and Select Create.
Once your project is created select Add to Project
and Visual Recognition Model
to add the Detect Object
custom model into your project.
You’ll then be prompted to associate a Visual Recognition service.
NOTE: As of Oct 2020, Watson Visual Recognition is no longer available in Lite plan.
If you have an existing Visual Recognition service, simply select the service from the list of available services.
If you don’t have a service available, select New from the top of the page and choose an appropriate plan. The Lite Plan is more than sufficient to get up and running with a proof of concept. If you plan on building out a beefier application then take a look at the Standard Plan.
Alright, that’s the boring stuff out of the way. You can now get to loading and training our model. Once you’ve associated your Visual Recognition service you’ll be redirected to the Collections page where you’re able to load images to train your model on.
Click Browse from the right hand side of the page to start uploading images.
In this case there’s a bunch of space shuttle images that will be used to train the model. These images are available from the GitHub repository inside the folder called IMAGES. Choose whichever images you’d like to use to build your own model.
Once you’ve selected the images to upload, they’ll start processing and eventually will be loaded into the project.
Once that’s done, you can start labelling your images!
The next part of this journey requires you to label the objects within your images. This is actually pretty straight forward inside of Watson Studio. All you have to do is select the image and drag a square around the object you’re trying to train on.
To add images to your training data, select the checkbox next to each image you’d like to use as part of the training set then select Add to Model.
This will load the images into the collection in preparation for training.
Once the image is loaded into the collection you’re able to start labelling it. Click into the image, then select Add new object.
You’ll then be able to draw squares (aka labels) over the objects within the image. Once you’ve selected Add new object all you have to do is draw a square over the object you’re labelling. In this case label the External Tank (Orange thing), the Boosters (white things next to the orange thing) and the Shuttle (the big white thing sitting on top of the others).
Once this has been done, write the name of the object you’re labelling. And hit Add. Repeat the same process for the other parts of the image. Once you’ve labelled all objects required. Select Done in the top right of the image labelled area.
To label the next set of images just hit the left and right arrows next to the current image to start labelling process again on the next image.
Anddd….you can label this image too!
Once you’ve labelled all your images, select View all images to see the results of your labelling process.
At this stage the deep learning model hasn’t been trained yet. To kick off this process select Train Model in the top right hand corner. This will commence the training process for your visual recognition model.
Once the training has finished, you’ll be able to test out your model with new images and view the results!
Now that you’ve labelled your images and trained the model you can then start the fun bit and commence testing your model. As soon as the model has finished training you’ll be redirected to an overview.
From here you can test out the model by selecting Test from the menu bar.
Then upload an image to test out your model.
The results of the detection will be visible once each image has been processed.