Hello, I am currently using the ML5 Neural Net to train my own image classifier and CNN. On the reference page for this method I see the inputs are image width, height and channels. Can anymore inputs be added and/or can more outputs than label and confidence be returned?
I want to also return a bounding box around the classified input. From what it looks like via the documentation this is currently not possible with the imageClassifier mode. However, if there is a way I can add more input options and include w,h,x,y in my training data that would be great. Or to calculate it on the output results.
As I understand so does the MLl5 NN and ML5 ImageClassifier go through every pixel on the image and do not identify anything that can be bounded in a box. These models just classify the features in all pixels based on similar label inputs. They return an array of the three labels with highest confidence score. Index 0 has the highest confidence score. They do not identify things in the image. If you train your model with cats on the right side of the image they will have problems classifying an image of an cat in the left side. To identify things there are models like ObjectDetection that kan classify what object are where in the image and then put a bounding box around them…I think
