I've seen a lot of people use image set classification recently. However, I haven't been able to find anywhere explaining the specifics of why image set classification is preferred. Are there only advantages to image set classification over normal image classification? Are there instances you might prefer the conventional image classification over image set classification, and are image set classification methods preferred for certain tasks/situations? Basically, what are the main differences between normal image classification and image set classification, and what are their advantages/disadvantages over the other?
1
-
$\begingroup$ Could you share what you mean by set classification? $\endgroup$– RoyiCommented May 19, 2023 at 10:36
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
As the name of the approach suggests, the data you train your model on is composed of multiple image sets. Each set contains images of the same label. For example, you can capture the same object from multiple angles and put those images into a single set, the idea being that the object X is the object X regardless of which angle you look at it. Such approach would probably be the right choice if you are trying to classify in real time while the objects are not static. As to whether you should always apply image set classification, I am yet to encounter an algorithm/method that has no drawbacks and is the correct choice regardless of the situation. Traditional image classification still has its place as it is likely to be less computationally expensive. Note that I am not an expert on the topic and I am just sharing what I know and some of my intuitions. You will find the paper Deep Bayesian Image Set Classification: A Defence Approach against Adversarial Attacks particularly helpful if you want to investigate the topic.
