DropLoss for Long-Tail Instance Segmentation

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. Codes are available at https://github.com/timy90022/DropLoss.

Improving the Intra-class Long-Tail in 3D Detection via Rare Example Mining

“…Long-tail visual recognition: Long-tail is conventionally defined as an imbalance in a multinomial distribution between various different class labels, either in the image classification context [8,24,26,27,36,55,62,64], dense segmentation problems [20,23,52,53,56,59], or between foreground / background labels in object detection problems [33,34,50,51,60]. Existing approaches for addressing class-imbalanced problems include resampling (oversampling tail classes or head classes), reweighitng (using inverse class frequency, effective number of samples [8]), novel loss function design [1,34,[50][51][52]63], meta learning for head-to-tail knowlege transfer [7,27,35,55], distillation [32,57] and mixture of experts [54].…”