Sejong University researchers examine hidden annotation errors in object detection datasets

The study systematically demonstrates how missing or erroneous labels in commonly used datasets negatively impact model performance, emphasizing that the reliability of object detection models depends on the quality of the training data.