A list of supported labels/classes for Object Detection model?

Question

I am using the Object Detection API for a detection task, but I am not sure if the model can actually detect the objects we want to detect. My questions are:

1. Is there a list of supported classes (i.e. all possible class labels) for the Object detection model that I can access? If not, how should I verify that the Object Detection API can detect a (quite long) list of objects I need?
2. Is there a list of supported tags for Image Description model?

Any help would be appreciated!

Answer 1

anonymous user-1557 The list of all possible object classes that the API can detect are not available to list through API directly but the models trained behind our computer vision API are trained to detect vast pool of objects which also lists the object hierarchy in the response. For example, here is a part of the response when analyze API detects objects in the image. This is possible by passing the optional visualFeatures parameter Objects.

"object": "Strawberry", "confidence": 0.588, "parent": { "object": "Fruit", "confidence": 0.612, "parent": { "object": "Food", "confidence": 0.62 } }  

In this response the object detected is fruit and its parent is food along with a confidence level. If you have any issues in getting your objects detected in an image you can provide feedback to us through this channel by providing the image or raise a support request from Azure portal so we could train our models to update this list.

For, the second part of your question the list of all available tags are also not available for public but you can pass the optional Tags parameter for visualFeatures and get the list for your image and let us know if it does not provide valid list of tags for your image.

Answer 2

@romungi-MSFT Can the list of all supported labels be provided?

1. We need to know the full list of supported labels in order to know if the service can detect the objects we are interested in.
2. Our use case need to check if our objects of interest are detected. Therefore we need to do a mapping on the supported labels to our long list of objects of interest.