P3M-10k

该数据集名为P3M-10k，是首个大规模的匿名化隐私保护人像抠图基准，包含了10,000张高分辨率的面部模糊人像图片以及高质量的alpha蒙版。该数据集用于在隐私保护训练设置下，评估无需三边图和基于三边图抠图方法的性能。规模达到了10,000张图像，任务专注于隐私保护人像抠图。

The dataset named P3M-10k is the first large-scale anonymized privacy-preserving portrait matting benchmark. It consists of 10,000 high-resolution face-blurred portrait images and high-quality alpha mattes. This benchmark is designed to evaluate the performance of both trimap-free and trimap-based matting methods under privacy-preserving training settings. With a total of 10,000 images, the dataset focuses on the task of privacy-preserving portrait matting.

搜集汇总

背景与挑战

背景概述

P3M-10k是首个大规模匿名化的人像抠图基准数据集，包含10,421张高分辨率面部模糊人像图像及高质量alpha遮罩，支持隐私保护训练（PPT）设置。数据集分为训练集（9,421张）和测试集（P3M-500-P和P3M-500-NP各500张），旨在评估模型在隐私保护场景下的性能。