Can custom vision measure object geometry from images? Since the face recognition service can identify and create polygons of face features, one may consider custom vision can do the same for specified features. Extract those polygons with some reference dimension units. The object of interest can be represented by the polygons and the dimensions can be approximated or estimated. Then use custom vision to quantify object size.

Please add the tensorflowjs model option.

It's so irritating not able to get an endpoint after train a model

I tagged some images and discovered that those tags are too specific. I would like to merge one or more tags. Do I need to label each image again? Couldn't I rename a tag to a name that already exists and group them?

If I try to rename a tag to another existing name, it shows me an error that this tag name already exists

When tagging and adjusting object location in image details, please retain the page the user is on after closing the image details modal dialog. As you go through all the images to tag or make adjustments to the object locations it is very frustrating to lose the page the you were on after closing the dialog.

Is there any way to version the model file? As we iterate, there does not seem to be anyway to identify the exported ONNX model by iteration, version, etc. To be clear I do not mean the Version used to create the model...but rather as we add images and retrain...the version of the result. The LearningModel class has a Version property but it seems to always return 0.

Object Detection learns actively and starts making region suggestions as the user goes along. Please please add the ability to suggest tags for Image Classification! This would really speed up our tagging scenarios.

I would like the ability to select a subset of images for Training, to enable my own cross-validation. Right now I don't think I have any control over which specific images are included in the model, unless I delete them (for later re-importing).

Since General Model has better accuracy, it should be possible to download model artefacts such as model in a .pb or .pkl file so that it can be deployed in a different environment or a different resource group.

It would be nice to be able to download all the photos I have uploaded in my custom vision.ai project.

I am using an Azure Function to push images into Custom Vision from Azure Blob Storage in order for it to be labelled via the GUI.

We would like to subscribe to an event stream/message queue for when a label has been applied/updated so that we can respond appropriately.

A mechanism for an Azure Function to trigger on this would be amazing.

Alternatively if we have to poll periodically we would like a way to identify which images have had labels applied/modified since the last iteration - is there a workaround to do this?

When using customvision.ai, we would like to pass images stored in a Blob storage to the prediction API.

I wish there was a way to use the exported model on a Google Coral Edge (or are there similar devices that can run the current models?). I converted the model to tensorflow lite but was not able to convert it to a Edge TPU Model - as I'm not an specialist on these topics, I don't really know what the issue is - but probably the format just does not match the requirements for Edge TPU Models.

When browsing predictions, the current page is changed back to first page after opening an image and assigning a tag to it.

Heat Map is very helpful to understand what area is impacted to Object Detetion task.
Hope we can see it at Active Learning phase via REST API call.