0
$\begingroup$

I'm aware that the ground-truth of the example at the top left-hand corner of the image below is "zero"

enter image description here

However, I am confused about the meaning of the terms ground truth and ground-truth labels. What is the difference between them?

Could someone give a hint of it?

$\endgroup$
1
$\begingroup$

These two terms could easily refer to the same thing, depending on the context. For example, a lazy person could easily say something like this

We compute the loss/error between the prediction (of the model) and the ground truth.

Here, the ground-truth refers to the "officially correct" label (categorical or numerical) for a given input with which you compute the prediction. So, in this case, ground-truth would be a synonym for a ground-truth label.

However, in general, ground-truth refers to anything, not just labels, that are correct or true (hence the name), so it could be used more generally. For instance, you could say something like this

We assume that the ground-truth underlying probability distribution from which the data is sampled is a Gaussian.

However, in this case, you could also leave out the ground-truth part, as it's more or less implied by the fact that you're assuming something.

So, the difference between the two is that "ground-truth" can be used more generally to refer to anything that is "true".

$\endgroup$
1
  • $\begingroup$ Note that this answer is based on my experience. I also don't think that these terms are really official, although they are often used in research papers, tutorials, etc. $\endgroup$
    – nbro
    Jun 5 at 13:07
2
$\begingroup$

Ground Truth

'Ground truth' is that data or information that you have that is 'true' or assumed to be true. That means that you have high or perfect knowledge of what it is. For example, in your image of numbers, you know that the first row are zeros, the second row are ones, the third are twos, and so on. You have 10 rows of data, each row is of a different class or category. Each class has 16 samples. Ground truth data is used to train machine learning or deep learning models. The example you provided is from the Modified National Institute of Standards and Technology (MNIST) database which is commonly used for building image classifiers for handwritten digits.

Ground Truth Labels

The 'ground-truth labels' are the names you choose to give them. You may choose to label the classes as '0', '1', '2', etc., or as 'zero', 'one', 'two', etc. Maybe you think in Greek. If so label them as 'μηδέν', 'ένα', 'δύο', etc.

Reference MNIST database

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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