question

NeilBarnwell-3329 avatar image
0 Votes"
NeilBarnwell-3329 asked romungi-MSFT answered

What's the correct tooling to use cognitive services to identity SPECIFIC items?

Scenario: I have three children and need to sort their garments as they are cleaned. I would like to be able to hold a garment up to a camera, and for the application to recognise the actual garment, so that it can tell me who it (currently) belongs to.

What approach should I use to teach an application to find a specific item? Obviously just noticing that it is a "t-shirt" or "trousers" isn't enough in this case; I need to identify individual items.

azure-machine-learningazure-cognitive-services
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

1 Answer

romungi-MSFT avatar image
0 Votes"
romungi-MSFT answered

@NeilBarnwell-3329 I would advise to use the Azure custom vision service for this scenario. Identifying an object in an image is possible by Azure computer vision and it does a good job to provide description of the image, tag all objects, locate the object, brands in an image with co-ordinates but it wouldn't be able to track the item with different predictions as the model is not customized for a closed set of objects.

Custom vision provides the ability to train a model with a closed set of objects and you would be able to predict an object in an image with high confidence and track the object with the help of tags. Tags are metadata that you would attach to an image while training so the trained model returns these tags with a confidence score. For example, I have trained a model with images of aircrafts to distinguish an image from a thumbnail. If the image contains a play button then it identifies the same in the tag or else the no_play tag is displayed.

If no play button is on the image.
140462-image.png

If play button is on the image.
140420-image.png

In your case, you would have to create a classification project with multilabel classification type for retail domain with appropriate tags set on your images to identify a particular garment or its owner. I hope this helps to get you started and try this scenario.


If an answer is helpful, please click on 130616-image.png or upvote 130671-image.png which might help other community members reading this thread.


image.png (399.3 KiB)
image.png (310.5 KiB)
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.