DropLoss for Long-Tail Instance Segmentation

TL;DR

DropLoss introduces an adaptive loss function that addresses the imbalance in long-tail instance segmentation by compensating for the suppression of rare categories, leading to improved performance across all category frequencies.

Contribution

The paper proposes DropLoss, a novel adaptive loss that effectively balances rare and frequent categories without trade-offs, advancing long-tail instance segmentation.

Findings

Achieves state-of-the-art mAP on LVIS dataset

Effectively balances rare and frequent categories

Improves detection accuracy for long-tail distributions

Abstract

Long-tailed class distributions are prevalent among the practical applications of object detection and instance segmentation. Prior work in long-tail instance segmentation addresses the imbalance of losses between rare and frequent categories by reducing the penalty for a model incorrectly predicting a rare class label. We demonstrate that the rare categories are heavily suppressed by correct background predictions, which reduces the probability for all foreground categories with equal weight. Due to the relative infrequency of rare categories, this leads to an imbalance that biases towards predicting more frequent categories. Based on this insight, we develop DropLoss -- a novel adaptive loss to compensate for this imbalance without a trade-off between rare and frequent categories. With this loss, we show state-of-the-art mAP across rare, common, and frequent categories on the LVIS dataset.