-1

I am trying to train a learning model to recognize which part (left/central/right) of a known object is represented in an image, assuming that the model's input can be one of the following: an image of the left part of the object (whole left part or some smaller portion of the left part); an image of its central part (whole central part or some smaller portion of the central part); an image of its right part (whole right part or some smaller portion of the right part). The position of the object is always fixed so all images are taken in front of the object and this will be also the case in which the model will be asked to make a prediction. I've collected a few thousands images belonging to the left, central and right part of the object; for each part, as explained, some images represent the whole part while others represent a smaller portion of that part, anyway i'm just interested to predict which part of the object (left/central/right) the image belongs to, so i've tackled this as a classification task over 3 labels but I'm wondering if the best choice for this task is to use a convolutional neural network or some other approach.

Any suggestion is very appreciated. Thanks

Carlo De Simone
  • 1

1 Answers1

0

Since you are trying to distinguish between three independent states, a classification approach is a sensible place to start. You could start with this tutorial and then, as you suggested, add some convolutional layers.

There are alternative approaches: A classification task normally means that each of the wrong answers is equally wrong. If the correct answer is 'left' and your algorithm gives the answer 'right', is this worse than giving the answer 'middle'? If so you might consider this as a regression problem.

Lastly, you may prefer to use the AI stack exchange forum for conceptual questions, as stackoverflow is normally reserved for specific coding questions.

jmsinusa
  • 1,584
  • 1
  • 13
  • 21