Abstract
Abstract
This thesis introduces a refined version of Average Precision Loss (AP-Loss), specifically designed to address class imbalance in long-tailed visual recognition. Recognizing the challenges posed by skewed class distributions in visual datasets, we have focused on modifying AP-Loss to address underrepresented classes better. Our research includes a comprehensive comparative analysis, where the adapted AP-Loss is assessed against standard loss functions across datasets with varying degrees of class imbalance. A critical aspect of our approach is the systematic evaluation of AP-Loss under different imbalance ratios, demonstrating its potential for superior performance in long-tailed visual recognition scenarios. This work represents a significant contribution to the ongoing research on class imbalance problems, offering a novel direction for addressing these challenges in visual recognition tasks and paving the way for future research in this critical area.