Hi!
I’m updating some of my demos for Microsoft Ignite and I found an amazing new feature in Custom Vision: Suggested Tags (see references). This feature is super useful in scenarios for automatic detection, like the parking lot demo. I’ll use the official documentation to describe this feature
When you tag images for a Custom Vision model, the service uses the latest trained iteration of the model to predict the labels of untagged images. It then shows these predictions as suggested tags, based on the selected confidence threshold and prediction uncertainty. You can then either confirm or change the suggestions, speeding up the process of manually tagging the images for training.
Label images faster with suggested tags
And, as usual, let’s use 2 images to describe this. Once I add a new image to my Custom Vision project, I can start to select object and tag them. However, if I already trained my project, I will also see the [Suggested object on] option.
With the default threshold value of 66%, the auto label feature does not detect any area. However, if I low the level, in example to 28%, it will automatically detect one of the parking slots: Slot 3. Once I’m happy with the suggested objects, I can confirm these objects and that’s it! Super easy.
This feature is amazing, and I’m looking forward to using it in real projects to see how much time saves in image labeling scenarios.
Bonus: Below you can see the before and after of the demo project. My daughter also decorated the new parking lot box, with some IronMan content. I’ll need to figure out how to connect this with my session speech!
Happy coding!
Greetings @ Toronto
El Bruno
References
- Label images faster with suggested tags https://docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/suggested-tags
- PhotoMania, https://photomania.net/editor
El Bruno 