Department of Computer Engineering

Long-tailed visual recognition, where certain classes contain far fewer samples than others, poses a critical challenge in computer vision applications, and addressing its complexities is essential for mirroring the real-world frequency distribution of classes. In this thesis, we explore and refine the Average Precision Loss (AP-Loss) approach to better manage the challenge of class imbalance inherent in these tasks. Initially, we found that the standard AP-Loss performs similarly to traditional loss functions like cross-entropy when dealing with uneven class distributions. By introducing two specific modifications to AP-Loss, we significantly improved the model's accuracy in identifying rare classes and its overall performance across all classes. We conducted thorough experiments to compare these improved AP-Loss versions with other top-performing loss functions in the field. Our findings show that our modified AP-Loss versions are competitive with state-of-the-art loss functions. This research contributes to the ongoing discussion on how to tackle class imbalance in visual recognition, offering new adjustments to AP-Loss that are promising for future studies in this important area.