I am working on a Design of Experiments to determine the image adjustment settings for a given collection of aircraft images to maximize the probabilities of correctly identifying the aircraft types using ImageIdentify.
The > Neat Example for ImageIdentify in the documentation shows the use of a category Entity["Word", "car"].
When I use Entity["Word", "aircraft"], the identification is by broad aircraft type. I would like the probabilities by the entries in AircraftData[] so that I could see the probability that the image is an F-15 Eagle, F/A-18 Hornet, etc.
ImageIdentify[f15image, Entity["Word", "aircraft"],10, "Probability"] Returns
<|Entity["Concept", "Interceptor::zff3s"] -> 0.74917, Entity["Concept", "AttackAircraft::548v5"] -> 0.755533, Entity["Concept", "StealthFighter::54q24"] -> 0.00636256, Entity["Concept", "MilitaryPlane::whx2r"] -> 0.75629, Entity["Concept", "Aeroplane::239dv"] -> 0.999771, Entity["Concept", "JumboJet::558sx"] -> 0.12094, Entity["Concept", "Floatplane::5wd4t"] -> 0.0944084, Entity["Concept", "Jetliner::p464v"] -> 0.0241423, Entity["Concept", "FlyingBoat::t75c7"] -> 0.00215588, Entity["Concept", "JetPlane::q8z95"] -> 0.145083|> I would like to use ImageIdentify with AircraftData[] or EntityList["Aircraft"] to see if it can identify specific aircraft types. None of my attempts to date have succeeded.
ImageIdentify[F15, Entity["Aircraft"], 10, "Probability"] Returns
Entity["Aircraft"] is not a valid object. I am working through the article on Resolving my Entity Crisis.
A hierarchical diagram that shows the relationships between the Entity... functions and terminology would be most helpful.
Do I need to create a custom entity store to make this possible?
ImageIdentifyuses a pretrained neural network to give predictions from a preset list of entities/concepts. If you want to include more of them, you have to train your own neural network with relevant image examples. $\endgroup$Entity...functions and terminology. This is not intuitive. I am sure that I will need to train my own neural network using Janes and other sources. What is here is neat to a point - but not to the level that I need. $\endgroup$Classifyon this set of data, and you will get a classifier which you can then use on other, unannotated images. $\endgroup$Classify. I may try to use theNameandImageproperties ofAircraftData[]. Barring that I will use a database of aircraft types. $\endgroup$