A tool for generating a custom vision projects. It will automatically find and download images of topics you imput, train the model and export a confusion matrix in CSV format.
You'll need to install Visual Studio to build and run this code.
- Sign in or create a new account on customvision.ai
- After login - click on cog icon on the right hand side and copy key values into the appropriate app settings
- Register for the free Bing Search service then copy the key value into the app settings
-
Get a Computer Vision API key to perform the smart thumbnails of downloaded images
-
Take a copy of SAMPLEApp.config rename it app.config and paste your keys into this file and set the size of your image set (~70% to train and 30% to test)
The app will create a project in your custom vision dashboard with a random GUID. Please note that each time you run the app, a new project will be created. Next the app reads all the tags in the tags.csv
file. The system searches Bing Images for each tag to prepare the source set of images for the classifier.
By default it downloads ten images for each tag. This is set in the app.config as sizeOfImageSet. Ideally you should change this around 100 images to get better results. We perform image augmentation steps to increase the size of the training dataset. After downloading the images, the app will then upload the photos into the customvision.ai project and tag them accordingly.
Now the model has enough data to be trained. The app will train the model and set the default iteration to the newly trained model.
The model is ready for testing so the app will try to test the model the results of which are exported as a confusion matrix into result.csv
.
You can check the quality of your model by reviewing the confusion matrix.
As you are running the app you should see output similar to this animation.