0
$\begingroup$

If I hypothetically want to train an AI to recognize enemy soldiers by the color of their armband (for example green armband), should I feed the AI with only green armband soldiers or should I also show it pictures of other color armband soldiers too?

The AI should specifically only detect the green armband soldiers and NOT any other color, there must be only one label of soldier in the system and that's the green armband soldier, there is only one threshold and that is for the green soldier AND NOT ANY OTHER.

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ "hypothetically" $\endgroup$
    – liakoyras
    Commented Jul 13, 2023 at 17:56

1 Answer 1

Reset to default
0
$\begingroup$

"Hypothetically" speaking if you only want to detect green armbands then yes you have to feed green armbands to the network. But this might lead to some miss-classifications. If the colour of the armband is lets say a different shade of green. In that case the model won't be robust enough to predict that.

I would suggest training your network with different colours as this will make your model robust. The model will learn to differentiate between different colours as well as different shades of a specific colour.

On the other hand if you only train your network with a single colour then the model will learn to detect whether that colour is present or not. It won't be able to differentiate between other colours/shades of colours. In other words it wont be robust which might negatively impact soldiers decision making in the filed.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Generally speaking, is it enough to train the AI model solely with images of men in military uniforms wearing green armbands, labeled as "enemy", for it to successfully identify such individuals as the enemy? Or is it also necessary to include images of civilians and own troops, labeled as "not-enemy", to ensure the AI does not mistakenly categorize them as enemies? $\endgroup$
    – Henno
    Commented Jul 14, 2023 at 6:28
  • $\begingroup$ If I use a single set of images (those of individuals with green armbands) and establish a high enough confidence level (in percentage) to ensure only actual enemies (men in military uniforms with green armbands), will the individuals be then correctly identified? How effective that approach would be compared to providing the AI with two categories of images? $\endgroup$
    – Henno
    Commented Jul 14, 2023 at 6:45
  • $\begingroup$ As you mentioned in your first comment if a civilian or a friendly wears a green armband and their clothes slightly resemble military uniform then yes the model will classify them as unfriendly. Hence you need to feed the model both the classes i.e. class A which is friendly and class B which is the enemy. That way the model will be much more robust $\endgroup$
    – spectre
    Commented Jul 14, 2023 at 7:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.