Custom Vision is a Cognitive Service used to create, process and deploy image classifiers based on the user images. The user needs to upload the images and label the images and the algorithm uses the labels to train the images.
- Create a new project by navigating to customvision.ai website.
- Upload at least more than 30 images each type of fruit and label them appropriately as shown below.
- Once the images are uploaded and labeled then click the “Train” button to train the model using these images.
- Choose the Training Type. In this case, select “Quick Training”
- Quick Training – Can used for quick iterations.
- Advanced Training – Use advanced training for improved performance, especially on challenging and fine-grained datasets. With advanced training, users can specify a compute time budget for training and Custom Vision will experimentally identify the best training and augmentation settings.
- Once Training type is chosen and select “Train”.
- Once training is complete, click on Publish.
- Provide the name for the model and the Azure Cognitive resource.
- Once published you can see the below output and the Prediction URL.
- For other settings click on “settings” icon as shown below.
- Now we can use the Prediction URL for testing.