The Watson Visual Recognition service can be trained to recognize your own custom classes in images. In this lab you'll learn how to train a Watson Visual Recognition custom classifier to recognition wedding scenes using 50 images of weddings to serve as positive examples and 50 images that are not weddings to serve as negative examples.
After creating the custom classifier, you'll test it with different images not used to train it to validate it's accuracy.
If you are not already signed up for the IBM Cloud, sign up here
1.2.1 From the IBM Cloud Dashboard click on Create resource
1.2.2 Select the Watson category project type and then click on Visual Recognition
1.2.3 Make sure the Lite plan is selected and then click Create
1.2.4 Select Service credentials at the left and then click on View credentials next to the credentials generated for your service instance
1.2.5 Click on the icon to copy the credentials to the clipboard and then save them in a text file on your Desktop (or some other convenient location). You'll need the api_key value multiple times.
In this section you'll create a custom classifier using 50 images of wedding scenes and 50 images of scenes that are not weddings to serve as negative examples. Training custome classifers can be done using the Watson Visual Recognition REST API but you'll use a web application that wraps those APIs to provide a more user friendly experience
2.1.1 In a new browser tab launch the tool http://ibm.biz/vr-tool
2.1.2 Click on API Key and enter the value of the api_key that is in your saved service credentials. Click on the arrow icon to continue.
2.2.1 Click on Create classifier
2.2.2 i) Name the classifier Weddings , ii) name the first class wedding and then iii) click on choose your files to upload the file weddings50.zip in the training_data sub folder of this project. iv) Click Create
2.2.3 click on choose your files in the tile for negative examples to upload the file notweddings50.zip in the training_data sub folder of this project.
2.2.4 Delete any unused tiles in the UI by clicking on the X in the top right of the tile
2.2.5 Click Create to create the classifier. After the training files are uploaded the new classifier will appear in the training state.
2.2.6 The training will take about 3-5 minutes. Now is a good time for a coffee break. When the new classifier is trained the state will change to ready
2.2.7 Highlight the new classifier's id and copy it to the clipboard. Paste it to the same file you used to save the service's api_key. You'll need this later.
Test the classifier with the GUI tool using single images and then test using 60 images via a standalone app. Both Python and Java are provided. Choose the one that you feel most comfortable with.
Python app requirements
- Python 3.5 or later.
Java app requirements:
- Java 1.7 or later JVM
- Gradle Build Tool
3.1.1 Click on choose your files in the tile for your new classifier. Select a file 0001.jpg from the test_data/wedding sub folder of this project. Verify that the file is classified as a wedding with a high confidence level ( approx 90%).
3.1.2 Click on choose your files in the tile for your new classifier. Select a file 0001.jpg from the test_data/notwedding sub folder of this project. Verify that the file is not classified as a wedding (confidence level 0)
3.2.1 Edit the file settings.py in the test_apps/python sub folder of this project. Put in the values of your api_key and clasifier_id that you save earlier
3.2.2 In a command prompt or terminal navigate to the test_apps/python sub folder of this project. Run the following command to install the dependencies
pip install -r requirements.txt
3.2.3 Run the following command to run the tester application
python watsonvr-wedding-tester.py
3.2.4 Verify the app runs without errors and the output looks like the following.
Classifying 60 test images ...
Finished classifying 10 images
Finished classifying 20 images
Finished classifying 30 images
Finished classifying 40 images
Finished classifying 50 images
Finished classifying 60 images
Number of files 60
True positives 28
True negatives 30
False positives 0
False negatives 2
Accuracy 96.67%
False negative list
test_data/wedding/0029.jpg
test_data/wedding/0023.jpg
3.2.5 Look at the files that were reported as false positives or false negatives and see if you can see why the classifier had problems with these particular images.
3.3.1 Edit the file settings.properties in the watson-vr-java-tester/src/main/resources sub folder of this project. Put in the values of your api_key and classifier_id that you saved earlier
3.3.2 In a command prompt or terminal navigate to the watson-vr-java-tester sub folder of this project. Run the following command to build the app
Linux/Mac
./gradlew build
Windows
gradle.bat build
3.3.3 Run the following command to run the tester application
Linux/Mac
./gradlew run
Windows
gradlew.bat run
3.3.4 (Optional) Run the following command to generate Eclipse artifacts so the project can be imported into Eclipse
Linux/Mac
./gradlew eclipse
Windows
gradlew.bat eclipse
Note: after running the command import this folder as an existing project into Eclipse
3.3.5 Verify the app runs without errors and the output looks like the following.
Finished classifying 60 images
Number of files 60
True positives 28
True negatives 30
False positives 0
False negatives 2
Accuracy 96.67
False negative list
/Users/[email protected]/build/watsonvr-wedding/test_data/wedding/0029.jpg
/Users/[email protected]/build/watsonvr-wedding/test_data/wedding/0023.jpg
BUILD SUCCESSFUL in 55s
3 actionable tasks: 1 executed, 2 up-to-date
3.3.6 Look at the files that were reported as false positives or false negatives and see if you can see why the classifier had problems with these particular images.
Congratulations ! You successfully created a classifier to detect wedding pictures. With just 50 examples and 50 negative examples your were able to quickly created a classifier that is approximately 97% accurate on some randomly selected test examples.