Closed shiffman closed 6 years ago
Yes, that's totally true. The ImageNet name was inherited from the imagenet example in deeplearn.js (https://deeplearnjs.org/demos/imagenet/) but I agree that it does not makes much sense to the end user.
I like let classifier = new ml5.ImageClassifier('ModelNet');
Great! I've added a new "API" label for discussions around function naming, etc.
Solved by #99
Woohoo!
I'm wondering what the best naming is for
ImageNet
andKNNImageClassifier
. The following to me is a little confusing:Is there a naming convention to distinguish between "image labeling" (it's a "cat"!) based on a pre-trained model and "image labeling" based on similarity to new images that you are feeding through the model (it's like that image you just showed me!)?
Would saying "Classifier" for both actually help? The "KNN" being the distinguishing aspect?
Much like
TransformNet
becomingStyleTransfer
I thinkImageNet
doesn't really say much to the end user. ImageNet is the database of images that the model was trained on so it is important, but maybe this could be more clearly stated in the documentation. Also, in theory couldn't we eventually have anImageClassifier
object with a model not trained with theImageNet
database?