P3M-10k (Privacy-Preserving Portrait Matting Dataset)

P3M-10k ，这是第一个大规模匿名的隐私保护肖像抠图基准。 P3M-10k 包含 10,000 张高分辨率面部模糊肖像图像以及高质量的 alpha 遮罩。我们在 P3M-10k 上系统地评估了无 trimap 和基于 trimap 的抠图方法，发现现有的抠图方法在遵循隐私保护训练 (PPT) 设置时显示出不同的泛化能力，?.?.，在面部模糊图像上进行训练并对任意图像进行测试。

P3M-10k is the first large-scale anonymous privacy-preserving portrait matting benchmark. It contains 10,000 high-resolution facial blurred portrait images along with high-quality alpha mattes. We systematically evaluated both trimap-free and trimap-based matting methods on P3M-10k, and found that existing matting methods exhibit varying generalization capabilities when following the Privacy-Preserving Training (PPT) setting, i.e., training on facial blurred images and testing on arbitrary images.

搜集汇总

背景与挑战

背景概述

P3M-10k是一个大规模隐私保护肖像抠图数据集，包含10,000张高分辨率面部模糊图像及对应的alpha遮罩。该数据集用于评估在隐私保护训练设置下肖像抠图方法的泛化性能。