I have found some time to compare IBM Watson's image recognition with Google's.
Google's Cloud Vision documentation can be found here: https://cloud.google.com/vision/
IBM Watson's Visual Recognition documentation can be found here: https://console.ng.bluemix.net/docs/services/VisualRecognition/VisualRecognition.html
Both platforms come with nice sample applications.
I have used the WatsonVisualRecognitionClient from https://github.com/watson-developer-cloud/java-sdk
And for Google I used LabelApp from https://github.com/GoogleCloudPlatform/cloud-vision.git
These are both written in Java.
Before you can use the samples, you need to get authorization.
For IBM you can the Visual Recognition service to your bluemix webapplication and retrieve the api_key from the credentials.
For Google you need to get credentials from https://console.cloud.google.com/apis/credentials
I have used some of my own images to see the outputs (JSON) of the two services
With these few samples it is hard to make a claim, but for every request it took Watson more time than Google.
On average 2.5 seconds for Google and 3.1 seconds for IBM
In the sample java client for Google I had the option to ask for 5 labels, I have not found this (yet) in the IBM sample. This is why the numbers of labels Watson returns varies in the table.
Training with own data
With Watson you can provide training data and Google also has a service ( https://cloud.google.com/ml/) but this one is in limited preview right now.
Should I get access to both and I find the time to train with my own data, I will follow this article up with those results.