PAT: Pixel-wise Adaptive Training for Long-tailed Segmentation

arXiv:2404.05393v1 Announce Type: new
Abstract: Beyond class frequency, we recognize the impact of class-wise relationships among various class-specific predictions and the imbalance in label masks on long-tailed segmentation learning. To address these challenges, we propose an innovative Pixel-wise Adaptive Training (PAT) technique tailored for long-tailed segmentation. PAT has two key features: 1) class-wise gradient magnitude homogenization, and 2) pixel-wise class-specific loss adaptation (PCLA). First, the class-wise gradient magnitude homogenization helps alleviate the imbalance among label masks by ensuring equal consideration …

abstract arxiv beyond challenges class cs.ai cs.cv features impact key masks pixel predictions relationships segmentation training type wise