It would be helpful to have some tools for boolean operations on tags - for example, if I have an "id" tag I'd like to select all non-id images, and perhaps combine them with other subsets. TOPCAT shows how it might go (see e.g. www.star.bris.ac.uk/~mbt/topcat/sun253/SubsetWindow.html).

Currently there's a rate limit if you send too many images in a short time (what's the exact limit by the way?). I'm working on a project that does not send lots of image - but it sends about 10 or 20 withing 2-3 seconds. I then only get results for a few of them and get "Rate limit is exceeded" for the others. Would it be possible to have a limit on a daily basis instead of "x per second"?

Prediction images appear in the web GUI, but I can't see a way to make them available from there for new training iterations (learn as you go). Is there a way - or could one be added?

It would be great to have access to more some more basic stats on the training process - learning curves, errors, ROCs, AUC values - the more the better.

Please add customvision to azure portal and allow login via adfs. All the other cognitive services stuff is there except this one.

Ability to upload images through an API rather than manually uploading the images through the portal.

We have thousands of images already tagged by our users, but uploading manually by a human is tedious work. We would love to be able to preselect them in our applications and send them in batches with our predefined tags.

It would be great to have the ability identify where in an image the item is. Ie if you train a model on detecting missing shingles on a roof, it would be great to show a bounding box around where the missing shingle is, in the image.

more then 50% (maybe ammount given in body content).. if i tag many images of flowers and few of them has bees and others different kind of butterflies. i dont need information of bees and butterflies species that are not in the picture.

It would be nice to have Python/PHP/Nodes/Curl wrappers like Watson Custom Visual Recognition. Any plan?

Allow predictions for multiple images (files) in a single request similar to what is supported for training.

i.e. predictionEndpoint.PredictImages(Guid predictionKey, List<Stream> imageStreams)

Currently, if you want to train your project, all tags need to have at least 5 images otherwise an error will be thrown and your entire project will not get trained.

Please allow an option to ignore tags with less that 5 images when training

It would be nice to have a separate prediction for each object in a picture. For example, if i have a pen and a pencil in the same picture, i would like to have probability for each object (list of tags), and probability of number of objects perhaps. Or if i have two pencils in the same pictures, etc.