I have been playing around with the google cloud vision API, namely the logo detection feature. Basically I want to determine if an image is a logo, so I run it through the API. However, I always get different results every time I run it. Sometimes the API classifies it as a logo, and sometimes it does not. Is there any explanation for this and possibly a way to improve the accuracy?
EDIT: I have just determined what the problem really is. I am trying to detect logos on remote images on a public facing website, and occasionally (but not all the time) the following error is returned:
I20180409-21:25:38.090(8)? error:
I20180409-21:25:38.091(8)? { details: [],
I20180409-21:25:38.091(8)? code: 13,
I20180409-21:25:38.091(8)? message: 'We can not access the URL
currently. Please download the content and pass it in.' },
What is the cause for this issue and is there a way around it?
I have fixed this issue by downloading the image as a base64 encoded image and then passing that into the google vision API instead of the url