What does "Ground-truth Labels" mean?

Table of Contents

Ground-truth labels are the true or correct answers used in research and testing. They serve as a standard to compare against when checking how well a model performs. For example, in action detection in videos, ground-truth labels would show what actions actually happen at different times.

These labels help researchers train models to recognize actions accurately. In dense multi-label situations, where many actions can happen at once, having clear ground-truth labels is even more important. They guide the model to learn which actions occur together and when.

In AI and machine learning, having good ground-truth labels improves results. When models are tested against these labels, researchers can see how close the model's guesses are to the actual actions, helping them make better models in the future.