RHD（Rendered Hand Pose）

RHD的全称是渲染Hand Pose，是一个关于手势估计的图像数据集。数据集包含41,258训练集和2,728测试集。
每个样本提供:
RGB图像 (320X320像素) 深度图 (320X320像素) 类的分割蒙版 (320X320像素): 人，背景，3类为每个手指和1类为每个手掌21个手键点和uv坐标在图像帧，世界框架中的xyz坐标和可见性指标固有相机矩阵K。

The full name of RHD is Rendered Hand Pose Dataset, which is an image dataset dedicated to hand pose estimation. The dataset contains 41,258 training samples and 2,728 test samples.
Each sample includes the following data:
- RGB images with a resolution of 320×320 pixels
- Depth maps with a resolution of 320×320 pixels
- Class-wise segmentation masks (320×320 pixels), covering categories including human, background, 3 categories for each finger, and 1 category for the palm
- UV coordinates of 21 hand keypoints in the image frame, XYZ coordinates of these keypoints in the world frame, visibility indicators, and the intrinsic camera matrix K

搜集汇总

背景与挑战

背景概述

RHD（Rendered Hand Pose）是一个专注于手势估计的图像数据集，包含41,258个训练样本和2,728个测试样本。每个样本提供RGB图像、深度图、分割蒙版以及21个手关键点的坐标和可见性指标，适用于手部姿态分析和3D重建任务。