Let's say I have a task with only 3 classes - 0, 1 and 2. I have built some different classifiers to make pixel-wise predictions for input images. But what would be a proper way to combine their results?

The only (and might be naive) way I've come up with is to make a pixel-wise majority vote.

However, I have read this paper(section III.D.2)) and they used "average, min, max and multiple operations". I guess "average" is equivalent to majority vote, but how could "min, max" work? (It seems unreasonable to "max the label", then what they might be maximizing ? probabilities? ) Also, what does "multiple" mean?

TL;DR:

1. What does min, max and multiple mean as ensemble methods in a semantic segmentation task ?

2. Are there any other good choices to combine multiple classifiers for a semantic segmentation task except majority vote ?

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.