In order to use Google Cloud Vision as an image labeling provider, we need a concept mapping from its internal entity IDs to Wikidata IDs.
The GCV labeling API returns an [[ https://cloud.google.com/vision/docs/reference/rest/v1/AnnotateImageResponse | AnnotateImageResponse ]] with each entity represented by an [[ https://cloud.google.com/vision/docs/reference/rest/v1/AnnotateImageResponse#EntityAnnotation | EntityAnnotation ]] object. According to the spec, entity IDs (`mid`s) are opaque; no mapping to an external knowledge base like Freebase or Wikidata is provided. (Opacity also implies that there's no guarantee that they will remain stable.)
It looks like the only way to create a mapping is to find the (current) GCV entity IDs for concepts of interest and use WDQS to map them to Wikidata IDs.
Other ideas?