Ohh, not seen AutoML before. But yes, that does look like custom trained AIs hosted in the cloud :)
Interesting. On 20 February 2018 at 17:04, bFlood <[email protected]> wrote: > only in alpha now but isn't this what AutoML is suppose to solve? (at > least the photo based portion, they mentioned there will be other AutoML > modules). You load your own images, label them and then use the Vision API > to access the results > > https://cloud.google.com/automl/ > > > > On Tuesday, February 20, 2018 at 6:55:18 AM UTC-5, barryhunter wrote: >> >> Technically no. You can't upload your own training dataset, and train the >> AI to produce *your *labels. >> >> Would need to deploy your own implementation - possibly via >> https://opensource.google.com/projects/tensorflow - deployed on GCE >> >> https://www.tensorflow.org/tutorials/image_recognition >> >> https://cloud.google.com/solutions/running-distributed-tenso >> rflow-on-compute-engine >> >> >> >> ... but have had reasonable success building a *second *layer on top of >> the API, to produce custom labels. Not sure what the proper name is, some >> sort of Meta-AI. >> >> >> Basically >> >> 1) Run all your existing images though the Cloud Vision API, and gather >> the feature labels it produces. >> >> 2) You now have a mapping of 'labels' to your own 'value' - although >> somewhat fuzzy. Will have a long list of labels for each value. Store these >> in a database >> >> 3) To 'search' images by your "value". first look at the image(s) with >> that value, and gather all the labels. THen look for the images with the >> most similar labels. >> >> 4) Can also do 'vision lookup' by uploading a image to the Cloud Vision >> API, and retrieveing a list of labels. Then search your database looking >> for the images with the most 'similar' labels. Take the 'value from the >> best match(es). >> >> >> It doesnt matter that what the labels Google AI is detecting on the >> images, even if nothing like your 'values', just that they similar >> 'looking'. Say you have a value of 'animal'. The vision api may detect >> those images as a mixture of differnt animals (cow/dog etc), but you have a >> mapping back and forth, via your 'training' dataset. >> >> Not an AI expert, so probably messing up the terminology, but the labels >> that Google's AI produces become almost like a set of hidden neurons on >> your *final *Meta-AI >> >> -- > You received this message because you are subscribed to the Google Groups > "Google App Engine" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/google-appengine. > To view this discussion on the web visit https://groups.google.com/d/ > msgid/google-appengine/9638c9f7-63d8-457b-a047- > 4752054fb96b%40googlegroups.com > <https://groups.google.com/d/msgid/google-appengine/9638c9f7-63d8-457b-a047-4752054fb96b%40googlegroups.com?utm_medium=email&utm_source=footer> > . > > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/google-appengine. To view this discussion on the web visit https://groups.google.com/d/msgid/google-appengine/CAJCAUu%2B_M7qwNzQ8zVGYYqtA_k%3DebuTFpoT940sFiJ6cvs_ogA%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
