Multi-level tail pixel cutmix and scale attention for long-tailed scene parsing

X Mj

Keywords: long tail, cutmix, scale attention, scene parsing

TL;DR: pixel distribution-based cutmix for long-tailed scene parsing

Abstract:

CutMix has been proposed to improve the image classication model robustness against input corruptions and its out-of-distribution
detection performances. In this work, we propose a new multi-level tail pixel cutmix which will cut regions according to the tail pixel distribution for improving the long-tailed Scene Parsing. After getting new CutMix-ed images, we furthermore apply a scale attention model with multi scale augmentation on these images to get a more eective representation on long-tailed pixel distribution.