Ground Truth

Ground Truth is a relative concept, which refers to the measurement value obtained by an existing and reliable measurement method as a benchmark relative to the measurement value obtained by a new measurement method. The field of machine learning borrows this concept, and the process of using the trained model to infer samples can be regarded as a broad measurement behavior.In the field of deep learning, Ground Truth (commonly used in English, meaning "ground truth" or "benchmark truth" in Chinese, simply understood as the true value) refers to the accurate labels or data used to train and evaluate models. Ground Truth is used to evaluate the performance of ML models as well as to train and validate models.

For example, if an ML model is being developed to classify images of animals, the ground truth would be the correct label for each image, such as "cat," "dog," or "bird." The model would be trained on a dataset containing images and their corresponding ground truth labels, and its performance would be evaluated based on how accurately it predicts the correct label for new, unseen images.

Particularly for large datasets, obtaining true labels can be a time-consuming and labor-intensive process. It often requires manually reviewing and annotating each case in the dataset, which can take a significant amount of time. In some cases, true labels can be established using automated methods, but these procedures can be less reliable and require more manual evaluation and correction.

Importance of Ground Truth for Machine Learning

Ground truth is a key element of machine learning because it provides a standard for comparing the performance of ML models. It is also a key element of many machine learning tasks, such as supervised learning, where models are trained and validated using true labels. In contrast, in unsupervised learning, models must learn to recognize patterns and correlations in data without any explicit instructions, as ground truth labels may not be accessible.

References

【1】https://encord.com/glossary/ground-truth-definition/

【2】https://zh.wikipedia.org/wiki/