Today I want to go through out of the box solutions provided from AWS, Azure and Google Cloud. I will use a photo that is not an easy pick to start with.
Let's start with AWS first, I am using AWS Rekognition to sample an image.
Without any customisation out of the box service returns below result.
Let's try Azure using Vision API, default does not tell you a lot and it was not a really quick start since you need to create cognito service and related resources.
Azure Cognitive Services Computer Vision API was not confident as it returned "colorfully painted cup" result with 25% confidence.
Now let's try Google Vision which actually allows you to upload an image and try API without creating account.
Now that is quite disappointing when looking at the object detection however if we check labels, results are way better.
Conclusion, it seems like Google actually does have higher confidence when labelling an image compared to AWS and Azure.